aio.c

/*
 *	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.
 */
#define pr_fmt(fmt) "%s: " fmt, __func__

#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/errno.h>
#include <linux/time.h>
#include <linux/aio_abi.h>
#include <linux/export.h>
#include <linux/syscalls.h>
#include <linux/backing-dev.h>
#include <linux/uio.h>

#include <linux/sched.h>
#include <linux/fs.h>
#include <linux/file.h>
#include <linux/mm.h>
#include <linux/mman.h>
#include <linux/mmu_context.h>
#include <linux/percpu.h>
#include <linux/slab.h>
#include <linux/timer.h>
#include <linux/aio.h>
#include <linux/highmem.h>
#include <linux/workqueue.h>
#include <linux/security.h>
#include <linux/eventfd.h>
#include <linux/blkdev.h>
#include <linux/compat.h>
#include <linux/migrate.h>
#include <linux/ramfs.h>
#include <linux/percpu-refcount.h>
#include <linux/mount.h>

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

#include "internal.h"

#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;	/* Written to by userland or under ring_lock
				 * mutex by aio_read_events_ring(). */
	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

struct kioctx_table {
	struct rcu_head	rcu;
	unsigned	nr;
	struct kioctx	*table[];
};

struct kioctx_cpu {
	unsigned		reqs_available;
};

struct ctx_rq_wait {
	struct completion comp;
	atomic_t count;
};

struct kioctx {
	struct percpu_ref	users;
	atomic_t		dead;

	struct percpu_ref	reqs;

	unsigned long		user_id;

	struct __percpu kioctx_cpu *cpu;

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

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

	unsigned long		mmap_base;
	unsigned long		mmap_size;

	struct page		**ring_pages;
	long			nr_pages;

	struct work_struct	free_work;

	/*
	 * signals when all in-flight requests are done
	 */
	struct ctx_rq_wait	*rq_wait;

	struct {
		/*
		 * 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.
		 *
		 * We batch accesses to it with a percpu version.
		 */
		atomic_t	reqs_available;
	} ____cacheline_aligned_in_smp;

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

	struct {
		struct mutex	ring_lock;
		wait_queue_head_t wait;
	} ____cacheline_aligned_in_smp;

	struct {
		unsigned	tail;
		unsigned	completed_events;
		spinlock_t	completion_lock;
	} ____cacheline_aligned_in_smp;

	struct page		*internal_pages[AIO_RING_PAGES];
	struct file		*aio_ring_file;

	unsigned		id;
};

/*
 * We use ki_cancel == KIOCB_CANCELLED to indicate that a kiocb has been either
 * cancelled or completed (this makes a certain amount of sense because
 * successful cancellation - io_cancel() - does deliver the completion to
 * userspace).
 *
 * And since most things don't implement kiocb cancellation and we'd really like
 * kiocb completion to be lockless when possible, we use ki_cancel to
 * synchronize cancellation and completion - we only set it to KIOCB_CANCELLED
 * with xchg() or cmpxchg(), see batch_complete_aio() and kiocb_cancel().
 */
#define KIOCB_CANCELLED		((void *) (~0ULL))

struct aio_kiocb {
	struct kiocb		common;

	struct kioctx		*ki_ctx;
	kiocb_cancel_fn		*ki_cancel;

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

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

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

/*------ sysctl variables----*/
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 */
/*----end sysctl variables---*/

static struct kmem_cache	*kiocb_cachep;
static struct kmem_cache	*kioctx_cachep;

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);
	if (IS_ERR(inode))
		return ERR_CAST(inode);

	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,
	};
	struct dentry *root = mount_pseudo(fs_type, "aio:", NULL, &ops,
					   AIO_RING_MAGIC);

	if (!IS_ERR(root))
		root->d_sb->s_iflags |= SB_I_NOEXEC;
	return root;
}

/* 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)
{
	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.");

	kiocb_cachep = KMEM_CACHE(aio_kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
	kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC);

	pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page));

	return 0;
}
__initcall(aio_setup);

static void put_aio_ring_file(struct kioctx *ctx)
{
	struct file *aio_ring_file = ctx->aio_ring_file;
	struct address_space *i_mapping;

	if (aio_ring_file) {
		truncate_setsize(aio_ring_file->f_inode, 0);

		/* Prevent further access to the kioctx from migratepages */
		i_mapping = aio_ring_file->f_inode->i_mapping;
		spin_lock(&i_mapping->private_lock);
		i_mapping->private_data = NULL;
		ctx->aio_ring_file = NULL;
		spin_unlock(&i_mapping->private_lock);

		fput(aio_ring_file);
	}
}

static void aio_free_ring(struct kioctx *ctx)
{
	int i;

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

	for (i = 0; i < ctx->nr_pages; i++) {
		struct page *page;
		pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i,
				page_count(ctx->ring_pages[i]));
		page = ctx->ring_pages[i];
		if (!page)
			continue;
		ctx->ring_pages[i] = NULL;
		put_page(page);
	}

	if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) {
		kfree(ctx->ring_pages);
		ctx->ring_pages = NULL;
	}
}

static int aio_ring_mremap(struct vm_area_struct *vma)
{
	struct file *file = vma->vm_file;
	struct mm_struct *mm = vma->vm_mm;
	struct kioctx_table *table;
	int i, res = -EINVAL;

	spin_lock(&mm->ioctx_lock);
	rcu_read_lock();
	table = rcu_dereference(mm->ioctx_table);
	for (i = 0; i < table->nr; i++) {
		struct kioctx *ctx;

		ctx = table->table[i];
		if (ctx && ctx->aio_ring_file == file) {
			if (!atomic_read(&ctx->dead)) {
				ctx->user_id = ctx->mmap_base = vma->vm_start;
				res = 0;
			}
			break;
		}
	}

	rcu_read_unlock();
	spin_unlock(&mm->ioctx_lock);
	return res;
}

static const struct vm_operations_struct aio_ring_vm_ops = {
	.mremap		= aio_ring_mremap,
#if IS_ENABLED(CONFIG_MMU)
	.fault		= filemap_fault,
	.map_pages	= filemap_map_pages,
	.page_mkwrite	= filemap_page_mkwrite,
#endif
};

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

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

#if IS_ENABLED(CONFIG_MIGRATION)
static int aio_migratepage(struct address_space *mapping, struct page *new,
			struct page *old, enum migrate_mode mode)
{
	struct kioctx *ctx;
	unsigned long flags;
	pgoff_t idx;
	int rc;

	rc = 0;

	/* mapping->private_lock here protects against the kioctx teardown.  */
	spin_lock(&mapping->private_lock);
	ctx = mapping->private_data;
	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;
	} else
		rc = -EINVAL;

	if (rc != 0)
		goto out_unlock;

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

	rc = migrate_page_move_mapping(mapping, new, old, NULL, mode, 1);
	if (rc != MIGRATEPAGE_SUCCESS) {
		put_page(new);
		goto out_unlock;
	}

	/* 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.
	 */
	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);

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

out_unlock:
	mutex_unlock(&ctx->ring_lock);
out:
	spin_unlock(&mapping->private_lock);
	return rc;
}
#endif

static const struct address_space_operations aio_ctx_aops = {
	.set_page_dirty = __set_page_dirty_no_writeback,
#if IS_ENABLED(CONFIG_MIGRATION)
	.migratepage	= aio_migratepage,
#endif
};

static int aio_setup_ring(struct kioctx *ctx)
{
	struct aio_ring *ring;
	unsigned nr_events = ctx->max_reqs;
	struct mm_struct *mm = current->mm;
	unsigned long size, unused;
	int nr_pages;
	int i;
	struct file *file;

	/* 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;

	nr_pages = PFN_UP(size);
	if (nr_pages < 0)
		return -EINVAL;

	file = aio_private_file(ctx, nr_pages);
	if (IS_ERR(file)) {
		ctx->aio_ring_file = NULL;
		return -ENOMEM;
	}

	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;
		}
	}

	for (i = 0; i < nr_pages; i++) {
		struct page *page;
		page = find_or_create_page(file->f_inode->i_mapping,
					   i, GFP_HIGHUSER | __GFP_ZERO);
		if (!page)
			break;
		pr_debug("pid(%d) page[%d]->count=%d\n",
			 current->pid, i, page_count(page));
		SetPageUptodate(page);
		unlock_page(page);

		ctx->ring_pages[i] = page;
	}
	ctx->nr_pages = i;

	if (unlikely(i != nr_pages)) {
		aio_free_ring(ctx);
		return -ENOMEM;
	}

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

	if (down_write_killable(&mm->mmap_sem)) {
		ctx->mmap_size = 0;
		aio_free_ring(ctx);
		return -EINTR;
	}

	ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size,
				       PROT_READ | PROT_WRITE,
				       MAP_SHARED, 0, &unused);
	up_write(&mm->mmap_sem);
	if (IS_ERR((void *)ctx->mmap_base)) {
		ctx->mmap_size = 0;
		aio_free_ring(ctx);
		return -ENOMEM;
	}

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

	ctx->user_id = ctx->mmap_base;
	ctx->nr_events = nr_events; /* trusted copy */

	ring = kmap_atomic(ctx->ring_pages[0]);
	ring->nr = nr_events;	/* user copy */
	ring->id = ~0U;
	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);
	kunmap_atomic(ring);
	flush_dcache_page(ctx->ring_pages[0]);

	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)

void kiocb_set_cancel_fn(struct kiocb *iocb, kiocb_cancel_fn *cancel)
{
	struct aio_kiocb *req = container_of(iocb, struct aio_kiocb, common);
	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);

static int kiocb_cancel(struct aio_kiocb *kiocb)
{
	kiocb_cancel_fn *old, *cancel;

	/*
	 * 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 -EINVAL;

		old = cancel;
		cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED);
	} while (cancel != old);

	return cancel(&kiocb->common);
}

static void free_ioctx(struct work_struct *work)
{
	struct kioctx *ctx = container_of(work, struct kioctx, free_work);

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

	aio_free_ring(ctx);
	free_percpu(ctx->cpu);
	percpu_ref_exit(&ctx->reqs);
	percpu_ref_exit(&ctx->users);
	kmem_cache_free(kioctx_cachep, ctx);
}

static void free_ioctx_reqs(struct percpu_ref *ref)
{
	struct kioctx *ctx = container_of(ref, struct kioctx, reqs);

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

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

/*
 * 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_users(struct percpu_ref *ref)
{
	struct kioctx *ctx = container_of(ref, struct kioctx, users);
	struct aio_kiocb *req;

	spin_lock_irq(&ctx->ctx_lock);

	while (!list_empty(&ctx->active_reqs)) {
		req = list_first_entry(&ctx->active_reqs,
				       struct aio_kiocb, ki_list);

		list_del_init(&req->ki_list);
		kiocb_cancel(req);
	}

	spin_unlock_irq(&ctx->ctx_lock);

	percpu_ref_kill(&ctx->reqs);
	percpu_ref_put(&ctx->reqs);
}

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);
	table = rcu_dereference_raw(mm->ioctx_table);

	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);

					/* While kioctx setup is in progress,
					 * we are protected from page migration
					 * changes ring_pages by ->ring_lock.
					 */
					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);
		old = rcu_dereference_raw(mm->ioctx_table);

		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;
		}
	}
}

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);
}

/* ioctx_alloc
 *	Allocates and initializes an ioctx.  Returns an ERR_PTR if it failed.
 */
static struct kioctx *ioctx_alloc(unsigned nr_events)
{
	struct mm_struct *mm = current->mm;
	struct kioctx *ctx;
	int err = -ENOMEM;

	/*
	 * 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;

	/* Prevent overflows */
	if (nr_events > (0x10000000U / sizeof(struct io_event))) {
		pr_debug("ENOMEM: nr_events too high\n");
		return ERR_PTR(-EINVAL);
	}

	if (!nr_events || (unsigned long)nr_events > (aio_max_nr * 2UL))
		return ERR_PTR(-EAGAIN);

	ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL);
	if (!ctx)
		return ERR_PTR(-ENOMEM);

	ctx->max_reqs = nr_events;

	spin_lock_init(&ctx->ctx_lock);
	spin_lock_init(&ctx->completion_lock);
	mutex_init(&ctx->ring_lock);
	/* Protect against page migration throughout kiotx setup by keeping
	 * the ring_lock mutex held until setup is complete. */
	mutex_lock(&ctx->ring_lock);
	init_waitqueue_head(&ctx->wait);

	INIT_LIST_HEAD(&ctx->active_reqs);

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

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

	ctx->cpu = alloc_percpu(struct kioctx_cpu);
	if (!ctx->cpu)
		goto err;

	err = aio_setup_ring(ctx);
	if (err < 0)
		goto err;

	atomic_set(&ctx->reqs_available, ctx->nr_events - 1);
	ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4);
	if (ctx->req_batch < 1)
		ctx->req_batch = 1;

	/* limit the number of system wide aios */
	spin_lock(&aio_nr_lock);
	if (aio_nr + nr_events > (aio_max_nr * 2UL) ||
	    aio_nr + nr_events < aio_nr) {
		spin_unlock(&aio_nr_lock);
		err = -EAGAIN;
		goto err_ctx;
	}
	aio_nr += ctx->max_reqs;
	spin_unlock(&aio_nr_lock);

	percpu_ref_get(&ctx->users);	/* io_setup() will drop this ref */
	percpu_ref_get(&ctx->reqs);	/* free_ioctx_users() will drop this */

	err = ioctx_add_table(ctx, mm);
	if (err)
		goto err_cleanup;

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

	pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n",
		 ctx, ctx->user_id, mm, ctx->nr_events);
	return ctx;

err_cleanup:
	aio_nr_sub(ctx->max_reqs);
err_ctx:
	atomic_set(&ctx->dead, 1);
	if (ctx->mmap_size)
		vm_munmap(ctx->mmap_base, ctx->mmap_size);
	aio_free_ring(ctx);
err:
	mutex_unlock(&ctx->ring_lock);
	free_percpu(ctx->cpu);
	percpu_ref_exit(&ctx->reqs);
	percpu_ref_exit(&ctx->users);
	kmem_cache_free(kioctx_cachep, ctx);
	pr_debug("error allocating ioctx %d\n", err);
	return ERR_PTR(err);
}

/* 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 int kill_ioctx(struct mm_struct *mm, struct kioctx *ctx,
		      struct ctx_rq_wait *wait)
{
	struct kioctx_table *table;

	spin_lock(&mm->ioctx_lock);
	if (atomic_xchg(&ctx->dead, 1)) {
		spin_unlock(&mm->ioctx_lock);
		return -EINVAL;
	}

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

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

	/*
	 * 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);

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

	ctx->rq_wait = wait;
	percpu_ref_kill(&ctx->users);
	return 0;
}

/*
 * 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.
 */
void exit_aio(struct mm_struct *mm)
{
	struct kioctx_table *table = rcu_dereference_raw(mm->ioctx_table);
	struct ctx_rq_wait wait;
	int i, skipped;

	if (!table)
		return;

	atomic_set(&wait.count, table->nr);
	init_completion(&wait.comp);

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

		if (!ctx) {
			skipped++;
			continue;
		}

		/*
		 * 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.
		 */
		ctx->mmap_size = 0;
		kill_ioctx(mm, ctx, &wait);
	}

	if (!atomic_sub_and_test(skipped, &wait.count)) {
		/* Wait until all IO for the context are done. */
		wait_for_completion(&wait.comp);
	}

	RCU_INIT_POINTER(mm->ioctx_table, NULL);
	kfree(table);
}

static void put_reqs_available(struct kioctx *ctx, unsigned nr)
{
	struct kioctx_cpu *kcpu;
	unsigned long flags;

	local_irq_save(flags);
	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);
	}

	local_irq_restore(flags);
}

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

	local_irq_save(flags);
	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:
	local_irq_restore(flags);
	return ret;
}

/* refill_reqs_available
 *	Updates the reqs_available reference counts used for tracking the
 *	number of free slots in the completion ring.  This can be called
 *	from aio_complete() (to optimistically update reqs_available) or
 *	from aio_get_req() (the we're out of events case).  It must be
 *	called holding ctx->completion_lock.
 */
static void refill_reqs_available(struct kioctx *ctx, unsigned head,
                                  unsigned tail)
{
	unsigned events_in_ring, completed;

	/* Clamp head since userland can write to it. */
	head %= ctx->nr_events;
	if (head <= tail)
		events_in_ring = tail - head;
	else
		events_in_ring = ctx->nr_events - (head - tail);

	completed = ctx->completed_events;
	if (events_in_ring < completed)
		completed -= events_in_ring;
	else
		completed = 0;

	if (!completed)
		return;

	ctx->completed_events -= completed;
	put_reqs_available(ctx, completed);
}

/* user_refill_reqs_available
 *	Called to refill reqs_available when aio_get_req() encounters an
 *	out of space in the completion ring.
 */
static void user_refill_reqs_available(struct kioctx *ctx)
{
	spin_lock_irq(&ctx->completion_lock);
	if (ctx->completed_events) {
		struct aio_ring *ring;
		unsigned head;

		/* Access of ring->head may race with aio_read_events_ring()
		 * here, but that's okay since whether we read the old version
		 * or the new version, and either will be valid.  The important
		 * part is that head cannot pass tail since we prevent
		 * aio_complete() from updating tail by holding
		 * ctx->completion_lock.  Even if head is invalid, the check
		 * against ctx->completed_events below will make sure we do the
		 * safe/right thing.
		 */
		ring = kmap_atomic(ctx->ring_pages[0]);
		head = ring->head;
		kunmap_atomic(ring);

		refill_reqs_available(ctx, head, ctx->tail);