aio.c 39.5 KB
Newer Older
L
Linus Torvalds 已提交
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.
 */
K
Kent Overstreet 已提交
11 12
#define pr_fmt(fmt) "%s: " fmt, __func__

L
Linus Torvalds 已提交
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>
L
Linus Torvalds 已提交
19
#include <linux/syscalls.h>
20
#include <linux/backing-dev.h>
21
#include <linux/uio.h>
L
Linus Torvalds 已提交
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>
K
Kent Overstreet 已提交
29
#include <linux/percpu.h>
L
Linus Torvalds 已提交
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>
J
Jeff Moyer 已提交
37
#include <linux/blkdev.h>
38
#include <linux/compat.h>
39 40
#include <linux/migrate.h>
#include <linux/ramfs.h>
K
Kent Overstreet 已提交
41
#include <linux/percpu-refcount.h>
42
#include <linux/mount.h>
L
Linus Torvalds 已提交
43 44 45 46

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

A
Al Viro 已提交
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[];
};

K
Kent Overstreet 已提交
76 77 78 79
struct kioctx_cpu {
	unsigned		reqs_available;
};

80
struct kioctx {
K
Kent Overstreet 已提交
81
	struct percpu_ref	users;
K
Kent Overstreet 已提交
82
	atomic_t		dead;
83

K
Kent Overstreet 已提交
84 85
	struct percpu_ref	reqs;

86 87
	unsigned long		user_id;

K
Kent Overstreet 已提交
88 89 90 91 92 93 94
	struct __percpu kioctx_cpu *cpu;

	/*
	 * For percpu reqs_available, number of slots we move to/from global
	 * counter at a time:
	 */
	unsigned		req_batch;
95 96 97 98
	/*
	 * This is what userspace passed to io_setup(), it's not used for
	 * anything but counting against the global max_reqs quota.
	 *
K
Kent Overstreet 已提交
99
	 * The real limit is nr_events - 1, which will be larger (see
100 101
	 * aio_setup_ring())
	 */
102 103
	unsigned		max_reqs;

K
Kent Overstreet 已提交
104 105
	/* Size of ringbuffer, in units of struct io_event */
	unsigned		nr_events;
106

K
Kent Overstreet 已提交
107 108 109 110 111 112
	unsigned long		mmap_base;
	unsigned long		mmap_size;

	struct page		**ring_pages;
	long			nr_pages;

K
Kent Overstreet 已提交
113
	struct work_struct	free_work;
114

115 116 117 118 119
	/*
	 * signals when all in-flight requests are done
	 */
	struct completion *requests_done;

120
	struct {
121 122 123 124 125
		/*
		 * This counts the number of available slots in the ringbuffer,
		 * so we avoid overflowing it: it's decremented (if positive)
		 * when allocating a kiocb and incremented when the resulting
		 * io_event is pulled off the ringbuffer.
K
Kent Overstreet 已提交
126 127
		 *
		 * We batch accesses to it with a percpu version.
128 129
		 */
		atomic_t	reqs_available;
130 131 132 133 134 135 136
	} ____cacheline_aligned_in_smp;

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

K
Kent Overstreet 已提交
137 138
	struct {
		struct mutex	ring_lock;
139 140
		wait_queue_head_t wait;
	} ____cacheline_aligned_in_smp;
K
Kent Overstreet 已提交
141 142 143 144

	struct {
		unsigned	tail;
		spinlock_t	completion_lock;
145
	} ____cacheline_aligned_in_smp;
K
Kent Overstreet 已提交
146 147

	struct page		*internal_pages[AIO_RING_PAGES];
148
	struct file		*aio_ring_file;
149 150

	unsigned		id;
151 152
};

L
Linus Torvalds 已提交
153
/*------ sysctl variables----*/
154 155 156
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 */
L
Linus Torvalds 已提交
157 158
/*----end sysctl variables---*/

159 160
static struct kmem_cache	*kiocb_cachep;
static struct kmem_cache	*kioctx_cachep;
L
Linus Torvalds 已提交
161

162 163 164 165 166 167 168 169 170 171 172
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);
173 174
	if (IS_ERR(inode))
		return ERR_CAST(inode);
175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203

	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,
	};
204
	return mount_pseudo(fs_type, "aio:", NULL, &ops, AIO_RING_MAGIC);
205 206
}

L
Linus Torvalds 已提交
207 208 209 210 211 212
/* 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)
{
213 214 215 216 217 218 219 220 221
	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.");

222 223
	kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
	kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC);
L
Linus Torvalds 已提交
224

K
Kent Overstreet 已提交
225
	pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page));
L
Linus Torvalds 已提交
226 227 228

	return 0;
}
229
__initcall(aio_setup);
L
Linus Torvalds 已提交
230

231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246
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);
	}
}

L
Linus Torvalds 已提交
247 248
static void aio_free_ring(struct kioctx *ctx)
{
249
	int i;
L
Linus Torvalds 已提交
250

251 252 253 254 255
	/* Disconnect the kiotx from the ring file.  This prevents future
	 * accesses to the kioctx from page migration.
	 */
	put_aio_ring_file(ctx);

256
	for (i = 0; i < ctx->nr_pages; i++) {
257
		struct page *page;
258 259
		pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i,
				page_count(ctx->ring_pages[i]));
260 261 262 263 264
		page = ctx->ring_pages[i];
		if (!page)
			continue;
		ctx->ring_pages[i] = NULL;
		put_page(page);
265
	}
L
Linus Torvalds 已提交
266

267
	if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) {
K
Kent Overstreet 已提交
268
		kfree(ctx->ring_pages);
269 270
		ctx->ring_pages = NULL;
	}
271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287
}

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;
}

288
#if IS_ENABLED(CONFIG_MIGRATION)
289 290 291
static int aio_migratepage(struct address_space *mapping, struct page *new,
			struct page *old, enum migrate_mode mode)
{
292
	struct kioctx *ctx;
293
	unsigned long flags;
294
	pgoff_t idx;
295 296
	int rc;

297 298
	rc = 0;

299
	/* mapping->private_lock here protects against the kioctx teardown.  */
300 301
	spin_lock(&mapping->private_lock);
	ctx = mapping->private_data;
302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320
	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;
321 322 323 324
	} else
		rc = -EINVAL;

	if (rc != 0)
325
		goto out_unlock;
326

327 328
	/* Writeback must be complete */
	BUG_ON(PageWriteback(old));
329
	get_page(new);
330

331
	rc = migrate_page_move_mapping(mapping, new, old, NULL, mode, 1);
332
	if (rc != MIGRATEPAGE_SUCCESS) {
333
		put_page(new);
334
		goto out_unlock;
335 336
	}

337 338 339
	/* 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.
340
	 */
341 342 343 344 345
	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);
346

347 348
	/* The old page is no longer accessible. */
	put_page(old);
349

350 351 352 353
out_unlock:
	mutex_unlock(&ctx->ring_lock);
out:
	spin_unlock(&mapping->private_lock);
354
	return rc;
L
Linus Torvalds 已提交
355
}
356
#endif
L
Linus Torvalds 已提交
357

358 359
static const struct address_space_operations aio_ctx_aops = {
	.set_page_dirty = aio_set_page_dirty,
360
#if IS_ENABLED(CONFIG_MIGRATION)
361
	.migratepage	= aio_migratepage,
362
#endif
363 364
};

L
Linus Torvalds 已提交
365 366 367 368
static int aio_setup_ring(struct kioctx *ctx)
{
	struct aio_ring *ring;
	unsigned nr_events = ctx->max_reqs;
Z
Zach Brown 已提交
369
	struct mm_struct *mm = current->mm;
370
	unsigned long size, unused;
L
Linus Torvalds 已提交
371
	int nr_pages;
372 373
	int i;
	struct file *file;
L
Linus Torvalds 已提交
374 375 376 377 378 379 380

	/* 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;

381
	nr_pages = PFN_UP(size);
L
Linus Torvalds 已提交
382 383 384
	if (nr_pages < 0)
		return -EINVAL;

385
	file = aio_private_file(ctx, nr_pages);
386 387
	if (IS_ERR(file)) {
		ctx->aio_ring_file = NULL;
388
		return -ENOMEM;
389 390
	}

391 392 393 394 395 396 397 398 399 400 401 402 403 404
	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;
		}
	}

405 406 407 408 409 410 411 412 413 414 415
	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);
416 417

		ctx->ring_pages[i] = page;
418
	}
419
	ctx->nr_pages = i;
L
Linus Torvalds 已提交
420

421 422
	if (unlikely(i != nr_pages)) {
		aio_free_ring(ctx);
423
		return -ENOMEM;
L
Linus Torvalds 已提交
424 425
	}

K
Kent Overstreet 已提交
426 427
	ctx->mmap_size = nr_pages * PAGE_SIZE;
	pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size);
428

Z
Zach Brown 已提交
429
	down_write(&mm->mmap_sem);
430 431
	ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size,
				       PROT_READ | PROT_WRITE,
432 433
				       MAP_SHARED, 0, &unused);
	up_write(&mm->mmap_sem);
K
Kent Overstreet 已提交
434 435
	if (IS_ERR((void *)ctx->mmap_base)) {
		ctx->mmap_size = 0;
L
Linus Torvalds 已提交
436
		aio_free_ring(ctx);
437
		return -ENOMEM;
L
Linus Torvalds 已提交
438 439
	}

K
Kent Overstreet 已提交
440
	pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base);
441

K
Kent Overstreet 已提交
442 443
	ctx->user_id = ctx->mmap_base;
	ctx->nr_events = nr_events; /* trusted copy */
L
Linus Torvalds 已提交
444

K
Kent Overstreet 已提交
445
	ring = kmap_atomic(ctx->ring_pages[0]);
L
Linus Torvalds 已提交
446
	ring->nr = nr_events;	/* user copy */
447
	ring->id = ~0U;
L
Linus Torvalds 已提交
448 449 450 451 452
	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);
453
	kunmap_atomic(ring);
K
Kent Overstreet 已提交
454
	flush_dcache_page(ctx->ring_pages[0]);
L
Linus Torvalds 已提交
455 456 457 458 459 460 461 462

	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)

463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478
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);

479
static int kiocb_cancel(struct kiocb *kiocb)
K
Kent Overstreet 已提交
480
{
481
	kiocb_cancel_fn *old, *cancel;
K
Kent Overstreet 已提交
482

483 484 485 486 487 488 489 490
	/*
	 * 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)
K
Kent Overstreet 已提交
491
			return -EINVAL;
K
Kent Overstreet 已提交
492

493 494 495
		old = cancel;
		cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED);
	} while (cancel != old);
K
Kent Overstreet 已提交
496

K
Kent Overstreet 已提交
497
	return cancel(kiocb);
K
Kent Overstreet 已提交
498 499
}

K
Kent Overstreet 已提交
500
static void free_ioctx(struct work_struct *work)
K
Kent Overstreet 已提交
501
{
K
Kent Overstreet 已提交
502
	struct kioctx *ctx = container_of(work, struct kioctx, free_work);
K
Kent Overstreet 已提交
503

K
Kent Overstreet 已提交
504
	pr_debug("freeing %p\n", ctx);
K
Kent Overstreet 已提交
505

K
Kent Overstreet 已提交
506
	aio_free_ring(ctx);
K
Kent Overstreet 已提交
507
	free_percpu(ctx->cpu);
508 509
	percpu_ref_exit(&ctx->reqs);
	percpu_ref_exit(&ctx->users);
K
Kent Overstreet 已提交
510 511 512
	kmem_cache_free(kioctx_cachep, ctx);
}

K
Kent Overstreet 已提交
513 514 515 516
static void free_ioctx_reqs(struct percpu_ref *ref)
{
	struct kioctx *ctx = container_of(ref, struct kioctx, reqs);

517 518 519 520
	/* At this point we know that there are no any in-flight requests */
	if (ctx->requests_done)
		complete(ctx->requests_done);

K
Kent Overstreet 已提交
521 522 523 524
	INIT_WORK(&ctx->free_work, free_ioctx);
	schedule_work(&ctx->free_work);
}

K
Kent Overstreet 已提交
525 526 527 528 529
/*
 * 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.
 */
K
Kent Overstreet 已提交
530
static void free_ioctx_users(struct percpu_ref *ref)
K
Kent Overstreet 已提交
531
{
K
Kent Overstreet 已提交
532
	struct kioctx *ctx = container_of(ref, struct kioctx, users);
K
Kent Overstreet 已提交
533 534 535 536 537 538 539 540 541
	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);
542
		kiocb_cancel(req);
K
Kent Overstreet 已提交
543 544 545 546
	}

	spin_unlock_irq(&ctx->ctx_lock);

K
Kent Overstreet 已提交
547 548
	percpu_ref_kill(&ctx->reqs);
	percpu_ref_put(&ctx->reqs);
K
Kent Overstreet 已提交
549 550
}

551 552 553 554 555 556 557
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);
558
	table = rcu_dereference_raw(mm->ioctx_table);
559 560 561 562 563 564 565 566 567

	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);

568 569 570 571
					/* While kioctx setup is in progress,
					 * we are protected from page migration
					 * changes ring_pages by ->ring_lock.
					 */
572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588
					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);
589
		old = rcu_dereference_raw(mm->ioctx_table);
590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605

		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;
		}
	}
}

K
Kent Overstreet 已提交
606 607 608 609 610 611 612 613 614 615
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);
}

L
Linus Torvalds 已提交
616 617 618 619 620
/* ioctx_alloc
 *	Allocates and initializes an ioctx.  Returns an ERR_PTR if it failed.
 */
static struct kioctx *ioctx_alloc(unsigned nr_events)
{
Z
Zach Brown 已提交
621
	struct mm_struct *mm = current->mm;
L
Linus Torvalds 已提交
622
	struct kioctx *ctx;
623
	int err = -ENOMEM;
L
Linus Torvalds 已提交
624

K
Kent Overstreet 已提交
625 626 627 628 629 630 631 632 633 634 635 636
	/*
	 * 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;

L
Linus Torvalds 已提交
637 638 639 640 641 642 643
	/* 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);
	}

644
	if (!nr_events || (unsigned long)nr_events > (aio_max_nr * 2UL))
L
Linus Torvalds 已提交
645 646
		return ERR_PTR(-EAGAIN);

647
	ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL);
L
Linus Torvalds 已提交
648 649 650 651 652 653
	if (!ctx)
		return ERR_PTR(-ENOMEM);

	ctx->max_reqs = nr_events;

	spin_lock_init(&ctx->ctx_lock);
654
	spin_lock_init(&ctx->completion_lock);
K
Kent Overstreet 已提交
655
	mutex_init(&ctx->ring_lock);
656 657 658
	/* Protect against page migration throughout kiotx setup by keeping
	 * the ring_lock mutex held until setup is complete. */
	mutex_lock(&ctx->ring_lock);
L
Linus Torvalds 已提交
659 660 661 662
	init_waitqueue_head(&ctx->wait);

	INIT_LIST_HEAD(&ctx->active_reqs);

663 664 665 666 667 668
	if (percpu_ref_init(&ctx->users, free_ioctx_users))
		goto err;

	if (percpu_ref_init(&ctx->reqs, free_ioctx_reqs))
		goto err;

K
Kent Overstreet 已提交
669 670
	ctx->cpu = alloc_percpu(struct kioctx_cpu);
	if (!ctx->cpu)
K
Kent Overstreet 已提交
671
		goto err;
L
Linus Torvalds 已提交
672

673 674
	err = aio_setup_ring(ctx);
	if (err < 0)
K
Kent Overstreet 已提交
675
		goto err;
K
Kent Overstreet 已提交
676

677
	atomic_set(&ctx->reqs_available, ctx->nr_events - 1);
K
Kent Overstreet 已提交
678
	ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4);
679 680
	if (ctx->req_batch < 1)
		ctx->req_batch = 1;
681

L
Linus Torvalds 已提交
682
	/* limit the number of system wide aios */
683
	spin_lock(&aio_nr_lock);
684
	if (aio_nr + nr_events > (aio_max_nr * 2UL) ||
685
	    aio_nr + nr_events < aio_nr) {
686
		spin_unlock(&aio_nr_lock);
K
Kent Overstreet 已提交
687
		err = -EAGAIN;
G
Gu Zheng 已提交
688
		goto err_ctx;
689 690
	}
	aio_nr += ctx->max_reqs;
691
	spin_unlock(&aio_nr_lock);
L
Linus Torvalds 已提交
692

693 694
	percpu_ref_get(&ctx->users);	/* io_setup() will drop this ref */
	percpu_ref_get(&ctx->reqs);	/* free_ioctx_users() will drop this */
K
Kent Overstreet 已提交
695

696 697
	err = ioctx_add_table(ctx, mm);
	if (err)
K
Kent Overstreet 已提交
698
		goto err_cleanup;
699

700 701 702
	/* Release the ring_lock mutex now that all setup is complete. */
	mutex_unlock(&ctx->ring_lock);

K
Kent Overstreet 已提交
703
	pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n",
K
Kent Overstreet 已提交
704
		 ctx, ctx->user_id, mm, ctx->nr_events);
L
Linus Torvalds 已提交
705 706
	return ctx;

K
Kent Overstreet 已提交
707 708
err_cleanup:
	aio_nr_sub(ctx->max_reqs);
G
Gu Zheng 已提交
709 710
err_ctx:
	aio_free_ring(ctx);
K
Kent Overstreet 已提交
711
err:
712
	mutex_unlock(&ctx->ring_lock);
K
Kent Overstreet 已提交
713
	free_percpu(ctx->cpu);
714 715
	percpu_ref_exit(&ctx->reqs);
	percpu_ref_exit(&ctx->users);
L
Linus Torvalds 已提交
716
	kmem_cache_free(kioctx_cachep, ctx);
K
Kent Overstreet 已提交
717
	pr_debug("error allocating ioctx %d\n", err);
718
	return ERR_PTR(err);
L
Linus Torvalds 已提交
719 720
}

K
Kent Overstreet 已提交
721 722 723 724 725
/* 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.
 */
726
static int kill_ioctx(struct mm_struct *mm, struct kioctx *ctx,
727
		struct completion *requests_done)
K
Kent Overstreet 已提交
728
{
729
	struct kioctx_table *table;
730

731 732
	if (atomic_xchg(&ctx->dead, 1))
		return -EINVAL;
733 734


735
	spin_lock(&mm->ioctx_lock);
736
	table = rcu_dereference_raw(mm->ioctx_table);
737 738 739
	WARN_ON(ctx != table->table[ctx->id]);
	table->table[ctx->id] = NULL;
	spin_unlock(&mm->ioctx_lock);
740

741 742
	/* percpu_ref_kill() will do the necessary call_rcu() */
	wake_up_all(&ctx->wait);
743

744 745 746 747 748 749 750 751
	/*
	 * 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);
752

753 754
	if (ctx->mmap_size)
		vm_munmap(ctx->mmap_base, ctx->mmap_size);
755

756 757 758
	ctx->requests_done = requests_done;
	percpu_ref_kill(&ctx->users);
	return 0;
L
Linus Torvalds 已提交
759 760 761 762 763
}

/* wait_on_sync_kiocb:
 *	Waits on the given sync kiocb to complete.
 */
K
Kent Overstreet 已提交
764
ssize_t wait_on_sync_kiocb(struct kiocb *req)
L
Linus Torvalds 已提交
765
{
K
Kent Overstreet 已提交
766
	while (!req->ki_ctx) {
L
Linus Torvalds 已提交
767
		set_current_state(TASK_UNINTERRUPTIBLE);
K
Kent Overstreet 已提交
768
		if (req->ki_ctx)
L
Linus Torvalds 已提交
769
			break;
J
Jeff Moyer 已提交
770
		io_schedule();
L
Linus Torvalds 已提交
771 772
	}
	__set_current_state(TASK_RUNNING);
K
Kent Overstreet 已提交
773
	return req->ki_user_data;
L
Linus Torvalds 已提交
774
}
775
EXPORT_SYMBOL(wait_on_sync_kiocb);
L
Linus Torvalds 已提交
776

K
Kent Overstreet 已提交
777 778 779 780 781 782 783
/*
 * 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.
L
Linus Torvalds 已提交
784
 */
785
void exit_aio(struct mm_struct *mm)
L
Linus Torvalds 已提交
786
{
787 788
	struct kioctx_table *table = rcu_dereference_raw(mm->ioctx_table);
	int i;
789

790 791
	if (!table)
		return;
792

793 794
	for (i = 0; i < table->nr; ++i) {
		struct kioctx *ctx = table->table[i];
J
Jens Axboe 已提交
795

796 797
		if (!ctx)
			continue;
798
		/*
799 800 801 802 803
		 * 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.
804
		 */
K
Kent Overstreet 已提交
805
		ctx->mmap_size = 0;
806
		kill_ioctx(mm, ctx, NULL);
L
Linus Torvalds 已提交
807
	}
808 809 810

	RCU_INIT_POINTER(mm->ioctx_table, NULL);
	kfree(table);
L
Linus Torvalds 已提交
811 812
}

K
Kent Overstreet 已提交
813 814 815
static void put_reqs_available(struct kioctx *ctx, unsigned nr)
{
	struct kioctx_cpu *kcpu;
816
	unsigned long flags;
K
Kent Overstreet 已提交
817

818
	local_irq_save(flags);
819
	kcpu = this_cpu_ptr(ctx->cpu);
K
Kent Overstreet 已提交
820
	kcpu->reqs_available += nr;
821

K
Kent Overstreet 已提交
822 823 824 825 826
	while (kcpu->reqs_available >= ctx->req_batch * 2) {
		kcpu->reqs_available -= ctx->req_batch;
		atomic_add(ctx->req_batch, &ctx->reqs_available);
	}

827
	local_irq_restore(flags);
K
Kent Overstreet 已提交
828 829 830 831 832 833
}

static bool get_reqs_available(struct kioctx *ctx)
{
	struct kioctx_cpu *kcpu;
	bool ret = false;
834
	unsigned long flags;
K
Kent Overstreet 已提交
835

836
	local_irq_save(flags);
837
	kcpu = this_cpu_ptr(ctx->cpu);
K
Kent Overstreet 已提交
838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855
	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:
856
	local_irq_restore(flags);
K
Kent Overstreet 已提交
857 858 859
	return ret;
}

L
Linus Torvalds 已提交
860
/* aio_get_req
K
Kent Overstreet 已提交
861 862
 *	Allocate a slot for an aio request.
 * Returns NULL if no requests are free.
L
Linus Torvalds 已提交
863
 */
K
Kent Overstreet 已提交
864
static inline struct kiocb *aio_get_req(struct kioctx *ctx)
L
Linus Torvalds 已提交
865
{
K
Kent Overstreet 已提交
866 867
	struct kiocb *req;

K
Kent Overstreet 已提交
868
	if (!get_reqs_available(ctx))
K
Kent Overstreet 已提交
869 870
		return NULL;

871
	req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO);
L
Linus Torvalds 已提交
872
	if (unlikely(!req))
K
Kent Overstreet 已提交
873
		goto out_put;
L
Linus Torvalds 已提交
874

K
Kent Overstreet 已提交
875 876
	percpu_ref_get(&ctx->reqs);

L
Linus Torvalds 已提交
877
	req->ki_ctx = ctx;
J
Jeff Moyer 已提交
878
	return req;
K
Kent Overstreet 已提交
879
out_put:
K
Kent Overstreet 已提交
880
	put_reqs_available(ctx, 1);
K
Kent Overstreet 已提交
881
	return NULL;
L
Linus Torvalds 已提交
882 883
}

884
static void kiocb_free(struct kiocb *req)
L
Linus Torvalds 已提交
885
{
886 887
	if (req->ki_filp)
		fput(req->ki_filp);
888 889
	if (req->ki_eventfd != NULL)
		eventfd_ctx_put(req->ki_eventfd);
L
Linus Torvalds 已提交
890 891 892
	kmem_cache_free(kiocb_cachep, req);
}

A
Adrian Bunk 已提交
893
static struct kioctx *lookup_ioctx(unsigned long ctx_id)
L
Linus Torvalds 已提交
894
{
895
	struct aio_ring __user *ring  = (void __user *)ctx_id;
J
Jens Axboe 已提交
896
	struct mm_struct *mm = current->mm;
897
	struct kioctx *ctx, *ret = NULL;
898 899 900 901 902
	struct kioctx_table *table;
	unsigned id;

	if (get_user(id, &ring->id))
		return NULL;
L
Linus Torvalds 已提交
903

J
Jens Axboe 已提交
904
	rcu_read_lock();
905
	table = rcu_dereference(mm->ioctx_table);
J
Jens Axboe 已提交
906

907 908
	if (!table || id >= table->nr)
		goto out;
L
Linus Torvalds 已提交
909

910
	ctx = table->table[id];
911
	if (ctx && ctx->user_id == ctx_id) {
912 913 914 915
		percpu_ref_get(&ctx->users);
		ret = ctx;
	}
out:
J
Jens Axboe 已提交
916
	rcu_read_unlock();
917
	return ret;
L
Linus Torvalds 已提交
918 919 920 921 922
}

/* aio_complete
 *	Called when the io request on the given iocb is complete.
 */
923
void aio_complete(struct kiocb *iocb, long res, long res2)
L
Linus Torvalds 已提交
924 925 926
{
	struct kioctx	*ctx = iocb->ki_ctx;
	struct aio_ring	*ring;
K
Kent Overstreet 已提交
927
	struct io_event	*ev_page, *event;
L
Linus Torvalds 已提交
928
	unsigned long	flags;
K
Kent Overstreet 已提交
929
	unsigned tail, pos;
L
Linus Torvalds 已提交
930

931 932 933 934 935 936
	/*
	 * 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
L
Linus Torvalds 已提交
937 938 939
	 */
	if (is_sync_kiocb(iocb)) {
		iocb->ki_user_data = res;
K
Kent Overstreet 已提交
940 941
		smp_wmb();
		iocb->ki_ctx = ERR_PTR(-EXDEV);
L
Linus Torvalds 已提交
942
		wake_up_process(iocb->ki_obj.tsk);
943
		return;
L
Linus Torvalds 已提交
944 945
	}

946 947 948 949 950 951 952
	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);
	}
953

954 955
	/*
	 * Add a completion event to the ring buffer. Must be done holding
956
	 * ctx->completion_lock to prevent other code from messing with the tail
957 958 959 960
	 * pointer since we might be called from irq context.
	 */
	spin_lock_irqsave(&ctx->completion_lock, flags);

K
Kent Overstreet 已提交
961
	tail = ctx->tail;
K
Kent Overstreet 已提交
962 963
	pos = tail + AIO_EVENTS_OFFSET;

K
Kent Overstreet 已提交
964
	if (++tail >= ctx->nr_events)
965
		tail = 0;
L
Linus Torvalds 已提交
966

K
Kent Overstreet 已提交
967
	ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
K
Kent Overstreet 已提交
968 969
	event = ev_page + pos % AIO_EVENTS_PER_PAGE;

L
Linus Torvalds 已提交
970 971 972 973 974
	event->obj = (u64)(unsigned long)iocb->ki_obj.user;
	event->data = iocb->ki_user_data;
	event->res = res;
	event->res2 = res2;

K
Kent Overstreet 已提交
975
	kunmap_atomic(ev_page);
K
Kent Overstreet 已提交
976
	flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
K
Kent Overstreet 已提交
977 978

	pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n",
K
Kent Overstreet 已提交
979 980
		 ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data,
		 res, res2);
L
Linus Torvalds 已提交
981 982 983 984 985 986

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

K
Kent Overstreet 已提交
987
	ctx->tail = tail;
L
Linus Torvalds 已提交
988

K
Kent Overstreet 已提交
989
	ring = kmap_atomic(ctx->ring_pages[0]);
K
Kent Overstreet 已提交
990
	ring->tail = tail;
991
	kunmap_atomic(ring);
K
Kent Overstreet 已提交
992
	flush_dcache_page(ctx->ring_pages[0]);
L
Linus Torvalds 已提交
993

994 995
	spin_unlock_irqrestore(&ctx->completion_lock, flags);

K
Kent Overstreet 已提交
996
	pr_debug("added to ring %p at [%u]\n", iocb, tail);
D
Davide Libenzi 已提交
997 998 999 1000 1001 1002

	/*
	 * 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.
	 */
1003
	if (iocb->ki_eventfd != NULL)
D
Davide Libenzi 已提交
1004 1005
		eventfd_signal(iocb->ki_eventfd, 1);

L
Linus Torvalds 已提交
1006
	/* everything turned out well, dispose of the aiocb. */
K
Kent Overstreet 已提交
1007
	kiocb_free(iocb);
1008
	put_reqs_available(ctx, 1);
L
Linus Torvalds 已提交
1009

1010 1011 1012 1013 1014 1015 1016 1017
	/*
	 * 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();

L
Linus Torvalds 已提交
1018 1019 1020
	if (waitqueue_active(&ctx->wait))
		wake_up(&ctx->wait);

K
Kent Overstreet 已提交
1021
	percpu_ref_put(&ctx->reqs);
L
Linus Torvalds 已提交
1022
}
1023
EXPORT_SYMBOL(aio_complete);
L
Linus Torvalds 已提交
1024

G
Gu Zheng 已提交
1025
/* aio_read_events_ring
1026 1027
 *	Pull an event off of the ioctx's event ring.  Returns the number of
 *	events fetched
L
Linus Torvalds 已提交
1028
 */
1029 1030
static long aio_read_events_ring(struct kioctx *ctx,
				 struct io_event __user *event, long nr)
L
Linus Torvalds 已提交
1031 1032
{
	struct aio_ring *ring;
1033
	unsigned head, tail, pos;
1034 1035 1036
	long ret = 0;
	int copy_ret;

K
Kent Overstreet 已提交
1037
	mutex_lock(&ctx->ring_lock);
L
Linus Torvalds 已提交
1038

1039
	/* Access to ->ring_pages here is protected by ctx->ring_lock. */
K
Kent Overstreet 已提交
1040
	ring = kmap_atomic(ctx->ring_pages[0]);
1041
	head = ring->head;
1042
	tail = ring->tail;
1043 1044
	kunmap_atomic(ring);

1045
	pr_debug("h%u t%u m%u\n", head, tail, ctx->nr_events);
L
Linus Torvalds 已提交
1046

1047
	if (head == tail)
L
Linus Torvalds 已提交
1048 1049
		goto out;

1050 1051 1052
	head %= ctx->nr_events;
	tail %= ctx->nr_events;

1053 1054 1055 1056 1057
	while (ret < nr) {
		long avail;
		struct io_event *ev;
		struct page *page;

1058 1059
		avail = (head <= tail ?  tail : ctx->nr_events) - head;
		if (head == tail)
1060 1061 1062 1063 1064 1065 1066
			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;
K
Kent Overstreet 已提交
1067
		page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE];
1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081
		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;
K
Kent Overstreet 已提交
1082
		head %= ctx->nr_events;
L
Linus Torvalds 已提交
1083 1084
	}

K
Kent Overstreet 已提交
1085
	ring = kmap_atomic(ctx->ring_pages[0]);
1086
	ring->head = head;
1087
	kunmap_atomic(ring);
K
Kent Overstreet 已提交
1088
	flush_dcache_page(ctx->ring_pages[0]);
1089

1090
	pr_debug("%li  h%u t%u\n", ret, head, tail);
1091
out:
K
Kent Overstreet 已提交
1092
	mutex_unlock(&ctx->ring_lock);
1093

L
Linus Torvalds 已提交
1094 1095 1096
	return ret;
}

1097 1098
static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr,
			    struct io_event __user *event, long *i)
L
Linus Torvalds 已提交
1099
{
1100
	long ret = aio_read_events_ring(ctx, event + *i, nr - *i);
L
Linus Torvalds 已提交
1101

1102 1103
	if (ret > 0)
		*i += ret;
L
Linus Torvalds 已提交
1104

1105 1106
	if (unlikely(atomic_read(&ctx->dead)))
		ret = -EINVAL;
L
Linus Torvalds 已提交
1107

1108 1109
	if (!*i)
		*i = ret;
L
Linus Torvalds 已提交
1110

1111
	return ret < 0 || *i >= min_nr;
L
Linus Torvalds 已提交
1112 1113
}

1114
static long read_events(struct kioctx *ctx, long min_nr, long nr,
L
Linus Torvalds 已提交
1115 1116 1117
			struct io_event __user *event,
			struct timespec __user *timeout)
{
1118 1119
	ktime_t until = { .tv64 = KTIME_MAX };
	long ret = 0;
L
Linus Torvalds 已提交
1120 1121 1122

	if (timeout) {
		struct timespec	ts;
1123

L
Linus Torvalds 已提交
1124
		if (unlikely(copy_from_user(&ts, timeout, sizeof(ts))))
1125
			return -EFAULT;
L
Linus Torvalds 已提交
1126

1127
		until = timespec_to_ktime(ts);
L
Linus Torvalds 已提交
1128 1129
	}

1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145
	/*
	 * 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);
L
Linus Torvalds 已提交
1146

1147 1148
	if (!ret && signal_pending(current))
		ret = -EINTR;
L
Linus Torvalds 已提交
1149

1150
	return ret;
L
Linus Torvalds 已提交
1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165
}

/* 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.
 */
1166
SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp)
L
Linus Torvalds 已提交
1167 1168 1169 1170 1171 1172 1173 1174 1175 1176
{
	struct kioctx *ioctx = NULL;
	unsigned long ctx;
	long ret;

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

	ret = -EINVAL;
1177 1178 1179
	if (unlikely(ctx || nr_events == 0)) {
		pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n",
		         ctx, nr_events);
L
Linus Torvalds 已提交
1180 1181 1182 1183 1184 1185 1186
		goto out;
	}

	ioctx = ioctx_alloc(nr_events);
	ret = PTR_ERR(ioctx);
	if (!IS_ERR(ioctx)) {
		ret = put_user(ioctx->user_id, ctxp);
1187
		if (ret)
1188
			kill_ioctx(current->mm, ioctx, NULL);
K
Kent Overstreet 已提交
1189
		percpu_ref_put(&ioctx->users);
L
Linus Torvalds 已提交
1190 1191 1192 1193 1194 1195 1196 1197 1198
	}

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
1199
 *	implemented.  May fail with -EINVAL if the context pointed to
L
Linus Torvalds 已提交
1200 1201
 *	is invalid.
 */
1202
SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx)
L
Linus Torvalds 已提交
1203 1204 1205
{
	struct kioctx *ioctx = lookup_ioctx(ctx);
	if (likely(NULL != ioctx)) {
1206 1207
		struct completion requests_done =
			COMPLETION_INITIALIZER_ONSTACK(requests_done);
1208
		int ret;
1209 1210 1211 1212 1213

		/* Pass requests_done to kill_ioctx() where it can be set
		 * in a thread-safe way. If we try to set it here then we have
		 * a race condition if two io_destroy() called simultaneously.
		 */
1214
		ret = kill_ioctx(current->mm, ioctx, &requests_done);
K
Kent Overstreet 已提交
1215
		percpu_ref_put(&ioctx->users);
1216 1217 1218 1219 1220

		/* Wait until all IO for the context are done. Otherwise kernel
		 * keep using user-space buffers even if user thinks the context
		 * is destroyed.
		 */
1221 1222
		if (!ret)
			wait_for_completion(&requests_done);
1223

1224
		return ret;
L
Linus Torvalds 已提交
1225 1226 1227 1228 1229
	}
	pr_debug("EINVAL: io_destroy: invalid context id\n");
	return -EINVAL;
}

K
Kent Overstreet 已提交
1230 1231
typedef ssize_t (aio_rw_op)(struct kiocb *, const struct iovec *,
			    unsigned long, loff_t);
1232
typedef ssize_t (rw_iter_op)(struct kiocb *, struct iov_iter *);
K
Kent Overstreet 已提交
1233

1234 1235 1236 1237 1238
static ssize_t aio_setup_vectored_rw(struct kiocb *kiocb,
				     int rw, char __user *buf,
				     unsigned long *nr_segs,
				     struct iovec **iovec,
				     bool compat)
B
Badari Pulavarty 已提交
1239 1240 1241
{
	ssize_t ret;

1242
	*nr_segs = kiocb->ki_nbytes;
K
Kent Overstreet 已提交
1243

1244 1245
#ifdef CONFIG_COMPAT
	if (compat)
K
Kent Overstreet 已提交
1246
		ret = compat_rw_copy_check_uvector(rw,
1247
				(struct compat_iovec __user *)buf,
1248
				*nr_segs, UIO_FASTIOV, *iovec, iovec);
1249 1250
	else
#endif
K
Kent Overstreet 已提交
1251
		ret = rw_copy_check_uvector(rw,
1252
				(struct iovec __user *)buf,
1253
				*nr_segs, UIO_FASTIOV, *iovec, iovec);
B
Badari Pulavarty 已提交
1254
	if (ret < 0)
K
Kent Overstreet 已提交
1255
		return ret;
1256

K
Kent Overstreet 已提交
1257
	/* ki_nbytes now reflect bytes instead of segs */
B
Badari Pulavarty 已提交
1258
	kiocb->ki_nbytes = ret;
K
Kent Overstreet 已提交
1259
	return 0;
B
Badari Pulavarty 已提交
1260 1261
}

1262 1263 1264 1265
static ssize_t aio_setup_single_vector(struct kiocb *kiocb,
				       int rw, char __user *buf,
				       unsigned long *nr_segs,
				       struct iovec *iovec)
B
Badari Pulavarty 已提交
1266
{
1267
	if (unlikely(!access_ok(!rw, buf, kiocb->ki_nbytes)))
K
Kent Overstreet 已提交
1268
		return -EFAULT;
1269

1270 1271 1272
	iovec->iov_base = buf;
	iovec->iov_len = kiocb->ki_nbytes;
	*nr_segs = 1;
B
Badari Pulavarty 已提交
1273 1274 1275
	return 0;
}

L
Linus Torvalds 已提交
1276
/*
G
Gu Zheng 已提交
1277 1278
 * aio_run_iocb:
 *	Performs the initial checks and io submission.
L
Linus Torvalds 已提交
1279
 */
1280 1281
static ssize_t aio_run_iocb(struct kiocb *req, unsigned opcode,
			    char __user *buf, bool compat)
L
Linus Torvalds 已提交
1282
{
K
Kent Overstreet 已提交
1283 1284
	struct file *file = req->ki_filp;
	ssize_t ret;
1285
	unsigned long nr_segs;
K
Kent Overstreet 已提交
1286 1287 1288
	int rw;
	fmode_t mode;
	aio_rw_op *rw_op;
1289
	rw_iter_op *iter_op;
1290
	struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
1291
	struct iov_iter iter;
L
Linus Torvalds 已提交
1292

1293
	switch (opcode) {
L
Linus Torvalds 已提交
1294
	case IOCB_CMD_PREAD:
B
Badari Pulavarty 已提交
1295
	case IOCB_CMD_PREADV:
K
Kent Overstreet 已提交
1296 1297 1298
		mode	= FMODE_READ;
		rw	= READ;
		rw_op	= file->f_op->aio_read;
1299
		iter_op	= file->f_op->read_iter;
K
Kent Overstreet 已提交
1300 1301 1302
		goto rw_common;

	case IOCB_CMD_PWRITE:
B
Badari Pulavarty 已提交
1303
	case IOCB_CMD_PWRITEV:
K
Kent Overstreet 已提交
1304 1305 1306
		mode	= FMODE_WRITE;
		rw	= WRITE;
		rw_op	= file->f_op->aio_write;
1307
		iter_op	= file->f_op->write_iter;
K
Kent Overstreet 已提交
1308 1309 1310 1311 1312
		goto rw_common;
rw_common:
		if (unlikely(!(file->f_mode & mode)))
			return -EBADF;

1313
		if (!rw_op && !iter_op)
K
Kent Overstreet 已提交
1314 1315
			return -EINVAL;

1316 1317 1318 1319 1320 1321
		ret = (opcode == IOCB_CMD_PREADV ||
		       opcode == IOCB_CMD_PWRITEV)
			? aio_setup_vectored_rw(req, rw, buf, &nr_segs,
						&iovec, compat)
			: aio_setup_single_vector(req, rw, buf, &nr_segs,
						  iovec);
1322 1323
		if (!ret)
			ret = rw_verify_area(rw, file, &req->ki_pos, req->ki_nbytes);
1324
		if (ret < 0) {
1325
			if (iovec != inline_vecs)
1326
				kfree(iovec);
K
Kent Overstreet 已提交
1327
			return ret;
1328
		}
K
Kent Overstreet 已提交
1329 1330 1331

		req->ki_nbytes = ret;

K
Kent Overstreet 已提交
1332 1333 1334 1335 1336 1337 1338 1339 1340 1341
		/* XXX: move/kill - rw_verify_area()? */
		/* This matches the pread()/pwrite() logic */
		if (req->ki_pos < 0) {
			ret = -EINVAL;
			break;
		}

		if (rw == WRITE)
			file_start_write(file);

1342 1343 1344 1345 1346 1347
		if (iter_op) {
			iov_iter_init(&iter, rw, iovec, nr_segs, req->ki_nbytes);
			ret = iter_op(req, &iter);
		} else {
			ret = rw_op(req, iovec, nr_segs, req->ki_pos);
		}
K
Kent Overstreet 已提交
1348 1349 1350

		if (rw == WRITE)
			file_end_write(file);
L
Linus Torvalds 已提交
1351
		break;
K
Kent Overstreet 已提交
1352

L
Linus Torvalds 已提交
1353
	case IOCB_CMD_FDSYNC:
K
Kent Overstreet 已提交
1354 1355 1356 1357
		if (!file->f_op->aio_fsync)
			return -EINVAL;

		ret = file->f_op->aio_fsync(req, 1);
L
Linus Torvalds 已提交
1358
		break;
K
Kent Overstreet 已提交
1359

L
Linus Torvalds 已提交
1360
	case IOCB_CMD_FSYNC:
K
Kent Overstreet 已提交
1361 1362 1363 1364
		if (!file->f_op->aio_fsync)
			return -EINVAL;

		ret = file->f_op->aio_fsync(req, 0);
L
Linus Torvalds 已提交
1365
		break;
K
Kent Overstreet 已提交
1366

L
Linus Torvalds 已提交
1367
	default:
K
Kent Overstreet 已提交
1368
		pr_debug("EINVAL: no operation provided\n");
K
Kent Overstreet 已提交
1369
		return -EINVAL;
L
Linus Torvalds 已提交
1370 1371
	}

1372
	if (iovec != inline_vecs)
1373 1374
		kfree(iovec);

K
Kent Overstreet 已提交
1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385
	if (ret != -EIOCBQUEUED) {
		/*
		 * There's no easy way to restart the syscall since other AIO's
		 * may be already running. Just fail this IO with EINTR.
		 */
		if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR ||
			     ret == -ERESTARTNOHAND ||
			     ret == -ERESTART_RESTARTBLOCK))
			ret = -EINTR;
		aio_complete(req, ret, 0);
	}
L
Linus Torvalds 已提交
1386 1387 1388 1389

	return 0;
}

A
Adrian Bunk 已提交
1390
static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
K
Kent Overstreet 已提交
1391
			 struct iocb *iocb, bool compat)
L
Linus Torvalds 已提交
1392 1393 1394 1395 1396
{
	struct kiocb *req;
	ssize_t ret;

	/* enforce forwards compatibility on users */
1397
	if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) {
K
Kent Overstreet 已提交
1398
		pr_debug("EINVAL: reserve field set\n");
L
Linus Torvalds 已提交
1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411
		return -EINVAL;
	}

	/* prevent overflows */
	if (unlikely(
	    (iocb->aio_buf != (unsigned long)iocb->aio_buf) ||
	    (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) ||
	    ((ssize_t)iocb->aio_nbytes < 0)
	   )) {
		pr_debug("EINVAL: io_submit: overflow check\n");
		return -EINVAL;
	}

K
Kent Overstreet 已提交
1412
	req = aio_get_req(ctx);
1413
	if (unlikely(!req))
L
Linus Torvalds 已提交
1414
		return -EAGAIN;
1415 1416 1417 1418 1419

	req->ki_filp = fget(iocb->aio_fildes);
	if (unlikely(!req->ki_filp)) {
		ret = -EBADF;
		goto out_put_req;
L
Linus Torvalds 已提交
1420
	}
1421

1422 1423 1424 1425 1426 1427 1428
	if (iocb->aio_flags & IOCB_FLAG_RESFD) {
		/*
		 * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an
		 * instance of the file* now. The file descriptor must be
		 * an eventfd() fd, and will be signaled for each completed
		 * event using the eventfd_signal() function.
		 */
1429
		req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd);
1430
		if (IS_ERR(req->ki_eventfd)) {
1431
			ret = PTR_ERR(req->ki_eventfd);
1432
			req->ki_eventfd = NULL;
1433 1434 1435
			goto out_put_req;
		}
	}
L
Linus Torvalds 已提交
1436

K
Kent Overstreet 已提交
1437
	ret = put_user(KIOCB_KEY, &user_iocb->aio_key);
L
Linus Torvalds 已提交
1438
	if (unlikely(ret)) {
K
Kent Overstreet 已提交
1439
		pr_debug("EFAULT: aio_key\n");
L
Linus Torvalds 已提交
1440 1441 1442 1443 1444 1445
		goto out_put_req;
	}

	req->ki_obj.user = user_iocb;
	req->ki_user_data = iocb->aio_data;
	req->ki_pos = iocb->aio_offset;
K
Kent Overstreet 已提交
1446
	req->ki_nbytes = iocb->aio_nbytes;
L
Linus Torvalds 已提交
1447

1448 1449 1450
	ret = aio_run_iocb(req, iocb->aio_lio_opcode,
			   (char __user *)(unsigned long)iocb->aio_buf,
			   compat);
Z
Zach Brown 已提交
1451
	if (ret)
1452
		goto out_put_req;
Z
Zach Brown 已提交
1453

L
Linus Torvalds 已提交
1454 1455
	return 0;
out_put_req:
K
Kent Overstreet 已提交
1456
	put_reqs_available(ctx, 1);
K
Kent Overstreet 已提交
1457
	percpu_ref_put(&ctx->reqs);
K
Kent Overstreet 已提交
1458
	kiocb_free(req);
L
Linus Torvalds 已提交
1459 1460 1461
	return ret;
}

1462 1463
long do_io_submit(aio_context_t ctx_id, long nr,
		  struct iocb __user *__user *iocbpp, bool compat)
L
Linus Torvalds 已提交
1464 1465 1466
{
	struct kioctx *ctx;
	long ret = 0;
J
Jeff Moyer 已提交
1467
	int i = 0;
S
Shaohua Li 已提交
1468
	struct blk_plug plug;
L
Linus Torvalds 已提交
1469 1470 1471 1472

	if (unlikely(nr < 0))
		return -EINVAL;

1473 1474 1475
	if (unlikely(nr > LONG_MAX/sizeof(*iocbpp)))
		nr = LONG_MAX/sizeof(*iocbpp);

L
Linus Torvalds 已提交
1476 1477 1478 1479 1480
	if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp)))))
		return -EFAULT;

	ctx = lookup_ioctx(ctx_id);
	if (unlikely(!ctx)) {
K
Kent Overstreet 已提交
1481
		pr_debug("EINVAL: invalid context id\n");
L
Linus Torvalds 已提交
1482 1483 1484
		return -EINVAL;
	}

S
Shaohua Li 已提交
1485 1486
	blk_start_plug(&plug);

L
Linus Torvalds 已提交
1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504
	/*
	 * AKPM: should this return a partial result if some of the IOs were
	 * successfully submitted?
	 */
	for (i=0; i<nr; i++) {
		struct iocb __user *user_iocb;
		struct iocb tmp;

		if (unlikely(__get_user(user_iocb, iocbpp + i))) {
			ret = -EFAULT;
			break;
		}

		if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) {
			ret = -EFAULT;
			break;
		}

K
Kent Overstreet 已提交
1505
		ret = io_submit_one(ctx, user_iocb, &tmp, compat);
L
Linus Torvalds 已提交
1506 1507 1508
		if (ret)
			break;
	}
S
Shaohua Li 已提交
1509
	blk_finish_plug(&plug);
L
Linus Torvalds 已提交
1510

K
Kent Overstreet 已提交
1511
	percpu_ref_put(&ctx->users);
L
Linus Torvalds 已提交
1512 1513 1514
	return i ? i : ret;
}

1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532
/* sys_io_submit:
 *	Queue the nr iocbs pointed to by iocbpp for processing.  Returns
 *	the number of iocbs queued.  May return -EINVAL if the aio_context
 *	specified by ctx_id is invalid, if nr is < 0, if the iocb at
 *	*iocbpp[0] is not properly initialized, if the operation specified
 *	is invalid for the file descriptor in the iocb.  May fail with
 *	-EFAULT if any of the data structures point to invalid data.  May
 *	fail with -EBADF if the file descriptor specified in the first
 *	iocb is invalid.  May fail with -EAGAIN if insufficient resources
 *	are available to queue any iocbs.  Will return 0 if nr is 0.  Will
 *	fail with -ENOSYS if not implemented.
 */
SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr,
		struct iocb __user * __user *, iocbpp)
{
	return do_io_submit(ctx_id, nr, iocbpp, 0);
}

L
Linus Torvalds 已提交
1533 1534 1535
/* lookup_kiocb
 *	Finds a given iocb for cancellation.
 */
1536 1537
static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb,
				  u32 key)
L
Linus Torvalds 已提交
1538 1539
{
	struct list_head *pos;
1540 1541 1542

	assert_spin_locked(&ctx->ctx_lock);

K
Kent Overstreet 已提交
1543 1544 1545
	if (key != KIOCB_KEY)
		return NULL;

L
Linus Torvalds 已提交
1546 1547 1548
	/* TODO: use a hash or array, this sucks. */
	list_for_each(pos, &ctx->active_reqs) {
		struct kiocb *kiocb = list_kiocb(pos);
K
Kent Overstreet 已提交
1549
		if (kiocb->ki_obj.user == iocb)
L
Linus Torvalds 已提交
1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564
			return kiocb;
	}
	return NULL;
}

/* sys_io_cancel:
 *	Attempts to cancel an iocb previously passed to io_submit.  If
 *	the operation is successfully cancelled, the resulting event is
 *	copied into the memory pointed to by result without being placed
 *	into the completion queue and 0 is returned.  May fail with
 *	-EFAULT if any of the data structures pointed to are invalid.
 *	May fail with -EINVAL if aio_context specified by ctx_id is
 *	invalid.  May fail with -EAGAIN if the iocb specified was not
 *	cancelled.  Will fail with -ENOSYS if not implemented.
 */
1565 1566
SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb,
		struct io_event __user *, result)
L
Linus Torvalds 已提交
1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581
{
	struct kioctx *ctx;
	struct kiocb *kiocb;
	u32 key;
	int ret;

	ret = get_user(key, &iocb->aio_key);
	if (unlikely(ret))
		return -EFAULT;

	ctx = lookup_ioctx(ctx_id);
	if (unlikely(!ctx))
		return -EINVAL;

	spin_lock_irq(&ctx->ctx_lock);
K
Kent Overstreet 已提交
1582

L
Linus Torvalds 已提交
1583
	kiocb = lookup_kiocb(ctx, iocb, key);
K
Kent Overstreet 已提交
1584
	if (kiocb)
1585
		ret = kiocb_cancel(kiocb);
K
Kent Overstreet 已提交
1586 1587 1588
	else
		ret = -EINVAL;

L
Linus Torvalds 已提交
1589 1590
	spin_unlock_irq(&ctx->ctx_lock);

K
Kent Overstreet 已提交
1591
	if (!ret) {
1592 1593 1594 1595
		/*
		 * The result argument is no longer used - the io_event is
		 * always delivered via the ring buffer. -EINPROGRESS indicates
		 * cancellation is progress:
K
Kent Overstreet 已提交
1596
		 */
1597
		ret = -EINPROGRESS;
K
Kent Overstreet 已提交
1598
	}
L
Linus Torvalds 已提交
1599

K
Kent Overstreet 已提交
1600
	percpu_ref_put(&ctx->users);
L
Linus Torvalds 已提交
1601 1602 1603 1604 1605 1606

	return ret;
}

/* io_getevents:
 *	Attempts to read at least min_nr events and up to nr events from
1607 1608 1609 1610 1611 1612 1613 1614
 *	the completion queue for the aio_context specified by ctx_id. If
 *	it succeeds, the number of read events is returned. May fail with
 *	-EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is
 *	out of range, if timeout is out of range.  May fail with -EFAULT
 *	if any of the memory specified is invalid.  May return 0 or
 *	< min_nr if the timeout specified by timeout has elapsed
 *	before sufficient events are available, where timeout == NULL
 *	specifies an infinite timeout. Note that the timeout pointed to by
J
Jeff Moyer 已提交
1615
 *	timeout is relative.  Will fail with -ENOSYS if not implemented.
L
Linus Torvalds 已提交
1616
 */
1617 1618 1619 1620 1621
SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id,
		long, min_nr,
		long, nr,
		struct io_event __user *, events,
		struct timespec __user *, timeout)
L
Linus Torvalds 已提交
1622 1623 1624 1625 1626
{
	struct kioctx *ioctx = lookup_ioctx(ctx_id);
	long ret = -EINVAL;

	if (likely(ioctx)) {
N
Namhyung Kim 已提交
1627
		if (likely(min_nr <= nr && min_nr >= 0))
L
Linus Torvalds 已提交
1628
			ret = read_events(ioctx, min_nr, nr, events, timeout);
K
Kent Overstreet 已提交
1629
		percpu_ref_put(&ioctx->users);
L
Linus Torvalds 已提交
1630 1631 1632
	}
	return ret;
}