[linux-2.6-block.git] / fs / iomap / direct-io.c

// SPDX-License-Identifier: GPL-2.0
/*
 * Copyright (C) 2010 Red Hat, Inc.
 * Copyright (c) 2016-2018 Christoph Hellwig.
 */
#include <linux/module.h>
#include <linux/compiler.h>
#include <linux/fs.h>
#include <linux/iomap.h>
#include <linux/backing-dev.h>
#include <linux/uio.h>
#include <linux/task_io_accounting_ops.h>

#include "../internal.h"

/*
 * Private flags for iomap_dio, must not overlap with the public ones in
 * iomap.h:
 */
#define IOMAP_DIO_WRITE_FUA	(1 << 28)
#define IOMAP_DIO_NEED_SYNC	(1 << 29)
#define IOMAP_DIO_WRITE		(1 << 30)
#define IOMAP_DIO_DIRTY		(1 << 31)

struct iomap_dio {
	struct kiocb		*iocb;
	const struct iomap_dio_ops *dops;
	loff_t			i_size;
	loff_t			size;
	atomic_t		ref;
	unsigned		flags;
	int			error;
	bool			wait_for_completion;

	union {
		/* used during submission and for synchronous completion: */
		struct {
			struct iov_iter		*iter;
			struct task_struct	*waiter;
			struct request_queue	*last_queue;
			blk_qc_t		cookie;
		} submit;

		/* used for aio completion: */
		struct {
			struct work_struct	work;
		} aio;
	};
};

int iomap_dio_iopoll(struct kiocb *kiocb, bool spin)
{
	struct request_queue *q = READ_ONCE(kiocb->private);

	if (!q)
		return 0;
	return blk_poll(q, READ_ONCE(kiocb->ki_cookie), spin);
}
EXPORT_SYMBOL_GPL(iomap_dio_iopoll);

static void iomap_dio_submit_bio(struct iomap_dio *dio, struct iomap *iomap,
		struct bio *bio)
{
	atomic_inc(&dio->ref);

	if (dio->iocb->ki_flags & IOCB_HIPRI)
		bio_set_polled(bio, dio->iocb);

	dio->submit.last_queue = bdev_get_queue(iomap->bdev);
	dio->submit.cookie = submit_bio(bio);
}

static ssize_t iomap_dio_complete(struct iomap_dio *dio)
{
	const struct iomap_dio_ops *dops = dio->dops;
	struct kiocb *iocb = dio->iocb;
	struct inode *inode = file_inode(iocb->ki_filp);
	loff_t offset = iocb->ki_pos;
	ssize_t ret = dio->error;

	if (dops && dops->end_io)
		ret = dops->end_io(iocb, dio->size, ret, dio->flags);

	if (likely(!ret)) {
		ret = dio->size;
		/* check for short read */
		if (offset + ret > dio->i_size &&
		    !(dio->flags & IOMAP_DIO_WRITE))
			ret = dio->i_size - offset;
		iocb->ki_pos += ret;
	}

	/*
	 * Try again to invalidate clean pages which might have been cached by
	 * non-direct readahead, or faulted in by get_user_pages() if the source
	 * of the write was an mmap'ed region of the file we're writing.  Either
	 * one is a pretty crazy thing to do, so we don't support it 100%.  If
	 * this invalidation fails, tough, the write still worked...
	 *
	 * And this page cache invalidation has to be after ->end_io(), as some
	 * filesystems convert unwritten extents to real allocations in
	 * ->end_io() when necessary, otherwise a racing buffer read would cache
	 * zeros from unwritten extents.
	 */
	if (!dio->error &&
	    (dio->flags & IOMAP_DIO_WRITE) && inode->i_mapping->nrpages) {
		int err;
		err = invalidate_inode_pages2_range(inode->i_mapping,
				offset >> PAGE_SHIFT,
				(offset + dio->size - 1) >> PAGE_SHIFT);
		if (err)
			dio_warn_stale_pagecache(iocb->ki_filp);
	}

	/*
	 * If this is a DSYNC write, make sure we push it to stable storage now
	 * that we've written data.
	 */
	if (ret > 0 && (dio->flags & IOMAP_DIO_NEED_SYNC))
		ret = generic_write_sync(iocb, ret);

	inode_dio_end(file_inode(iocb->ki_filp));
	kfree(dio);

	return ret;
}

static void iomap_dio_complete_work(struct work_struct *work)
{
	struct iomap_dio *dio = container_of(work, struct iomap_dio, aio.work);
	struct kiocb *iocb = dio->iocb;

	iocb->ki_complete(iocb, iomap_dio_complete(dio), 0);
}

/*
 * Set an error in the dio if none is set yet.  We have to use cmpxchg
 * as the submission context and the completion context(s) can race to
 * update the error.
 */
static inline void iomap_dio_set_error(struct iomap_dio *dio, int ret)
{
	cmpxchg(&dio->error, 0, ret);
}

static void iomap_dio_bio_end_io(struct bio *bio)
{
	struct iomap_dio *dio = bio->bi_private;
	bool should_dirty = (dio->flags & IOMAP_DIO_DIRTY);

	if (bio->bi_status)
		iomap_dio_set_error(dio, blk_status_to_errno(bio->bi_status));

	if (atomic_dec_and_test(&dio->ref)) {
		if (dio->wait_for_completion) {
			struct task_struct *waiter = dio->submit.waiter;
			WRITE_ONCE(dio->submit.waiter, NULL);
			blk_wake_io_task(waiter);
		} else if (dio->flags & IOMAP_DIO_WRITE) {
			struct inode *inode = file_inode(dio->iocb->ki_filp);

			INIT_WORK(&dio->aio.work, iomap_dio_complete_work);
			queue_work(inode->i_sb->s_dio_done_wq, &dio->aio.work);
		} else {
			iomap_dio_complete_work(&dio->aio.work);
		}
	}

	if (should_dirty) {
		bio_check_pages_dirty(bio);
	} else {
		bio_release_pages(bio, false);
		bio_put(bio);
	}
}

static void
iomap_dio_zero(struct iomap_dio *dio, struct iomap *iomap, loff_t pos,
		unsigned len)
{
	struct page *page = ZERO_PAGE(0);
	int flags = REQ_SYNC | REQ_IDLE;
	struct bio *bio;

	bio = bio_alloc(GFP_KERNEL, 1);
	bio_set_dev(bio, iomap->bdev);
	bio->bi_iter.bi_sector = iomap_sector(iomap, pos);
	bio->bi_private = dio;
	bio->bi_end_io = iomap_dio_bio_end_io;

	get_page(page);
	__bio_add_page(bio, page, len, 0);
	bio_set_op_attrs(bio, REQ_OP_WRITE, flags);
	iomap_dio_submit_bio(dio, iomap, bio);
}

static loff_t
iomap_dio_bio_actor(struct inode *inode, loff_t pos, loff_t length,
		struct iomap_dio *dio, struct iomap *iomap)
{
	unsigned int blkbits = blksize_bits(bdev_logical_block_size(iomap->bdev));
	unsigned int fs_block_size = i_blocksize(inode), pad;
	unsigned int align = iov_iter_alignment(dio->submit.iter);
	struct iov_iter iter;
	struct bio *bio;
	bool need_zeroout = false;
	bool use_fua = false;
	int nr_pages, ret = 0;
	size_t copied = 0;

	if ((pos | length | align) & ((1 << blkbits) - 1))
		return -EINVAL;

	if (iomap->type == IOMAP_UNWRITTEN) {
		dio->flags |= IOMAP_DIO_UNWRITTEN;
		need_zeroout = true;
	}

	if (iomap->flags & IOMAP_F_SHARED)
		dio->flags |= IOMAP_DIO_COW;

	if (iomap->flags & IOMAP_F_NEW) {
		need_zeroout = true;
	} else if (iomap->type == IOMAP_MAPPED) {
		/*
		 * Use a FUA write if we need datasync semantics, this is a pure
		 * data IO that doesn't require any metadata updates (including
		 * after IO completion such as unwritten extent conversion) and
		 * the underlying device supports FUA. This allows us to avoid
		 * cache flushes on IO completion.
		 */
		if (!(iomap->flags & (IOMAP_F_SHARED|IOMAP_F_DIRTY)) &&
		    (dio->flags & IOMAP_DIO_WRITE_FUA) &&
		    blk_queue_fua(bdev_get_queue(iomap->bdev)))
			use_fua = true;
	}

	/*
	 * Operate on a partial iter trimmed to the extent we were called for.
	 * We'll update the iter in the dio once we're done with this extent.
	 */
	iter = *dio->submit.iter;
	iov_iter_truncate(&iter, length);

	nr_pages = iov_iter_npages(&iter, BIO_MAX_PAGES);
	if (nr_pages <= 0)
		return nr_pages;

	if (need_zeroout) {
		/* zero out from the start of the block to the write offset */
		pad = pos & (fs_block_size - 1);
		if (pad)
			iomap_dio_zero(dio, iomap, pos - pad, pad);
	}

	do {
		size_t n;
		if (dio->error) {
			iov_iter_revert(dio->submit.iter, copied);
			return 0;
		}

		bio = bio_alloc(GFP_KERNEL, nr_pages);
		bio_set_dev(bio, iomap->bdev);
		bio->bi_iter.bi_sector = iomap_sector(iomap, pos);
		bio->bi_write_hint = dio->iocb->ki_hint;
		bio->bi_ioprio = dio->iocb->ki_ioprio;
		bio->bi_private = dio;
		bio->bi_end_io = iomap_dio_bio_end_io;

		ret = bio_iov_iter_get_pages(bio, &iter);
		if (unlikely(ret)) {
			/*
			 * We have to stop part way through an IO. We must fall
			 * through to the sub-block tail zeroing here, otherwise
			 * this short IO may expose stale data in the tail of
			 * the block we haven't written data to.
			 */
			bio_put(bio);
			goto zero_tail;
		}

		n = bio->bi_iter.bi_size;
		if (dio->flags & IOMAP_DIO_WRITE) {
			bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
			if (use_fua)
				bio->bi_opf |= REQ_FUA;
			else
				dio->flags &= ~IOMAP_DIO_WRITE_FUA;
			task_io_account_write(n);
		} else {
			bio->bi_opf = REQ_OP_READ;
			if (dio->flags & IOMAP_DIO_DIRTY)
				bio_set_pages_dirty(bio);
		}

		iov_iter_advance(dio->submit.iter, n);

		dio->size += n;
		pos += n;
		copied += n;

		nr_pages = iov_iter_npages(&iter, BIO_MAX_PAGES);
		iomap_dio_submit_bio(dio, iomap, bio);
	} while (nr_pages);

	/*
	 * We need to zeroout the tail of a sub-block write if the extent type
	 * requires zeroing or the write extends beyond EOF. If we don't zero
	 * the block tail in the latter case, we can expose stale data via mmap
	 * reads of the EOF block.
	 */
zero_tail:
	if (need_zeroout ||
	    ((dio->flags & IOMAP_DIO_WRITE) && pos >= i_size_read(inode))) {
		/* zero out from the end of the write to the end of the block */
		pad = pos & (fs_block_size - 1);
		if (pad)
			iomap_dio_zero(dio, iomap, pos, fs_block_size - pad);
	}
	return copied ? copied : ret;
}

static loff_t
iomap_dio_hole_actor(loff_t length, struct iomap_dio *dio)
{
	length = iov_iter_zero(length, dio->submit.iter);
	dio->size += length;
	return length;
}

static loff_t
iomap_dio_inline_actor(struct inode *inode, loff_t pos, loff_t length,
		struct iomap_dio *dio, struct iomap *iomap)
{
	struct iov_iter *iter = dio->submit.iter;
	size_t copied;

	BUG_ON(pos + length > PAGE_SIZE - offset_in_page(iomap->inline_data));

	if (dio->flags & IOMAP_DIO_WRITE) {
		loff_t size = inode->i_size;

		if (pos > size)
			memset(iomap->inline_data + size, 0, pos - size);
		copied = copy_from_iter(iomap->inline_data + pos, length, iter);
		if (copied) {
			if (pos + copied > size)
				i_size_write(inode, pos + copied);
			mark_inode_dirty(inode);
		}
	} else {
		copied = copy_to_iter(iomap->inline_data + pos, length, iter);
	}
	dio->size += copied;
	return copied;
}

static loff_t
iomap_dio_actor(struct inode *inode, loff_t pos, loff_t length,
		void *data, struct iomap *iomap, struct iomap *srcmap)
{
	struct iomap_dio *dio = data;

	switch (iomap->type) {
	case IOMAP_HOLE:
		if (WARN_ON_ONCE(dio->flags & IOMAP_DIO_WRITE))
			return -EIO;
		return iomap_dio_hole_actor(length, dio);
	case IOMAP_UNWRITTEN:
		if (!(dio->flags & IOMAP_DIO_WRITE))
			return iomap_dio_hole_actor(length, dio);
		return iomap_dio_bio_actor(inode, pos, length, dio, iomap);
	case IOMAP_MAPPED:
		return iomap_dio_bio_actor(inode, pos, length, dio, iomap);
	case IOMAP_INLINE:
		return iomap_dio_inline_actor(inode, pos, length, dio, iomap);
	default:
		WARN_ON_ONCE(1);
		return -EIO;
	}
}

/*
 * iomap_dio_rw() always completes O_[D]SYNC writes regardless of whether the IO
 * is being issued as AIO or not.  This allows us to optimise pure data writes
 * to use REQ_FUA rather than requiring generic_write_sync() to issue a
 * REQ_FLUSH post write. This is slightly tricky because a single request here
 * can be mapped into multiple disjoint IOs and only a subset of the IOs issued
 * may be pure data writes. In that case, we still need to do a full data sync
 * completion.
 */
ssize_t
iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,
		const struct iomap_ops *ops, const struct iomap_dio_ops *dops,
		bool wait_for_completion)
{
	struct address_space *mapping = iocb->ki_filp->f_mapping;
	struct inode *inode = file_inode(iocb->ki_filp);
	size_t count = iov_iter_count(iter);
	loff_t pos = iocb->ki_pos, start = pos;
	loff_t end = iocb->ki_pos + count - 1, ret = 0;
	unsigned int flags = IOMAP_DIRECT;
	struct blk_plug plug;
	struct iomap_dio *dio;

	lockdep_assert_held(&inode->i_rwsem);

	if (!count)
		return 0;

	if (WARN_ON(is_sync_kiocb(iocb) && !wait_for_completion))
		return -EIO;

	dio = kmalloc(sizeof(*dio), GFP_KERNEL);
	if (!dio)
		return -ENOMEM;

	dio->iocb = iocb;
	atomic_set(&dio->ref, 1);
	dio->size = 0;
	dio->i_size = i_size_read(inode);
	dio->dops = dops;
	dio->error = 0;
	dio->flags = 0;

	dio->submit.iter = iter;
	dio->submit.waiter = current;
	dio->submit.cookie = BLK_QC_T_NONE;
	dio->submit.last_queue = NULL;

	if (iov_iter_rw(iter) == READ) {
		if (pos >= dio->i_size)
			goto out_free_dio;

		if (iter_is_iovec(iter))
			dio->flags |= IOMAP_DIO_DIRTY;
	} else {
		flags |= IOMAP_WRITE;
		dio->flags |= IOMAP_DIO_WRITE;

		/* for data sync or sync, we need sync completion processing */
		if (iocb->ki_flags & IOCB_DSYNC)
			dio->flags |= IOMAP_DIO_NEED_SYNC;

		/*
		 * For datasync only writes, we optimistically try using FUA for
		 * this IO.  Any non-FUA write that occurs will clear this flag,
		 * hence we know before completion whether a cache flush is
		 * necessary.
		 */
		if ((iocb->ki_flags & (IOCB_DSYNC | IOCB_SYNC)) == IOCB_DSYNC)
			dio->flags |= IOMAP_DIO_WRITE_FUA;
	}

	if (iocb->ki_flags & IOCB_NOWAIT) {
		if (filemap_range_has_page(mapping, start, end)) {
			ret = -EAGAIN;
			goto out_free_dio;
		}
		flags |= IOMAP_NOWAIT;
	}

	ret = filemap_write_and_wait_range(mapping, start, end);
	if (ret)
		goto out_free_dio;

	/*
	 * Try to invalidate cache pages for the range we're direct
	 * writing.  If this invalidation fails, tough, the write will
	 * still work, but racing two incompatible write paths is a
	 * pretty crazy thing to do, so we don't support it 100%.
	 */
	ret = invalidate_inode_pages2_range(mapping,
			start >> PAGE_SHIFT, end >> PAGE_SHIFT);
	if (ret)
		dio_warn_stale_pagecache(iocb->ki_filp);
	ret = 0;

	if (iov_iter_rw(iter) == WRITE && !wait_for_completion &&
	    !inode->i_sb->s_dio_done_wq) {
		ret = sb_init_dio_done_wq(inode->i_sb);
		if (ret < 0)
			goto out_free_dio;
	}

	inode_dio_begin(inode);

	blk_start_plug(&plug);
	do {
		ret = iomap_apply(inode, pos, count, flags, ops, dio,
				iomap_dio_actor);
		if (ret <= 0) {
			/* magic error code to fall back to buffered I/O */
			if (ret == -ENOTBLK) {
				wait_for_completion = true;
				ret = 0;
			}
			break;
		}
		pos += ret;

		if (iov_iter_rw(iter) == READ && pos >= dio->i_size)
			break;
	} while ((count = iov_iter_count(iter)) > 0);
	blk_finish_plug(&plug);

	if (ret < 0)
		iomap_dio_set_error(dio, ret);

	/*
	 * If all the writes we issued were FUA, we don't need to flush the
	 * cache on IO completion. Clear the sync flag for this case.
	 */
	if (dio->flags & IOMAP_DIO_WRITE_FUA)
		dio->flags &= ~IOMAP_DIO_NEED_SYNC;

	WRITE_ONCE(iocb->ki_cookie, dio->submit.cookie);
	WRITE_ONCE(iocb->private, dio->submit.last_queue);

	/*
	 * We are about to drop our additional submission reference, which
	 * might be the last reference to the dio.  There are three three
	 * different ways we can progress here:
	 *
	 *  (a) If this is the last reference we will always complete and free
	 *	the dio ourselves.
	 *  (b) If this is not the last reference, and we serve an asynchronous
	 *	iocb, we must never touch the dio after the decrement, the
	 *	I/O completion handler will complete and free it.
	 *  (c) If this is not the last reference, but we serve a synchronous
	 *	iocb, the I/O completion handler will wake us up on the drop
	 *	of the final reference, and we will complete and free it here
	 *	after we got woken by the I/O completion handler.
	 */
	dio->wait_for_completion = wait_for_completion;
	if (!atomic_dec_and_test(&dio->ref)) {
		if (!wait_for_completion)
			return -EIOCBQUEUED;

		for (;;) {
			set_current_state(TASK_UNINTERRUPTIBLE);
			if (!READ_ONCE(dio->submit.waiter))
				break;

			if (!(iocb->ki_flags & IOCB_HIPRI) ||
			    !dio->submit.last_queue ||
			    !blk_poll(dio->submit.last_queue,
					 dio->submit.cookie, true))
				io_schedule();
		}
		__set_current_state(TASK_RUNNING);
	}

	return iomap_dio_complete(dio);

out_free_dio:
	kfree(dio);
	return ret;
}
EXPORT_SYMBOL_GPL(iomap_dio_rw);
Commit	Line	Data
db074436 DW	1	// SPDX-License-Identifier: GPL-2.0
	2	/*
	3	* Copyright (C) 2010 Red Hat, Inc.
	4	* Copyright (c) 2016-2018 Christoph Hellwig.
	5	*/
	6	#include <linux/module.h>
	7	#include <linux/compiler.h>
	8	#include <linux/fs.h>
	9	#include <linux/iomap.h>
	10	#include <linux/backing-dev.h>
	11	#include <linux/uio.h>
	12	#include <linux/task_io_accounting_ops.h>
	13
	14	#include "../internal.h"
	15
	16	/*
	17	* Private flags for iomap_dio, must not overlap with the public ones in
	18	* iomap.h:
	19	*/
	20	#define IOMAP_DIO_WRITE_FUA (1 << 28)
	21	#define IOMAP_DIO_NEED_SYNC (1 << 29)
	22	#define IOMAP_DIO_WRITE (1 << 30)
	23	#define IOMAP_DIO_DIRTY (1 << 31)
	24
	25	struct iomap_dio {
	26	struct kiocb *iocb;
838c4f3d	27	const struct iomap_dio_ops *dops;
db074436 DW	28	loff_t i_size;
	29	loff_t size;
	30	atomic_t ref;
	31	unsigned flags;
	32	int error;
	33	bool wait_for_completion;
	34
	35	union {
	36	/* used during submission and for synchronous completion: */
	37	struct {
	38	struct iov_iter *iter;
	39	struct task_struct *waiter;
	40	struct request_queue *last_queue;
	41	blk_qc_t cookie;
	42	} submit;
	43
	44	/* used for aio completion: */
	45	struct {
	46	struct work_struct work;
	47	} aio;
	48	};
	49	};
	50
	51	int iomap_dio_iopoll(struct kiocb *kiocb, bool spin)
	52	{
	53	struct request_queue *q = READ_ONCE(kiocb->private);
	54
	55	if (!q)
	56	return 0;
	57	return blk_poll(q, READ_ONCE(kiocb->ki_cookie), spin);
	58	}
	59	EXPORT_SYMBOL_GPL(iomap_dio_iopoll);
	60
	61	static void iomap_dio_submit_bio(struct iomap_dio dio, struct iomap iomap,
	62	struct bio *bio)
	63	{
	64	atomic_inc(&dio->ref);
	65
	66	if (dio->iocb->ki_flags & IOCB_HIPRI)
	67	bio_set_polled(bio, dio->iocb);
	68
	69	dio->submit.last_queue = bdev_get_queue(iomap->bdev);
	70	dio->submit.cookie = submit_bio(bio);
	71	}
	72
	73	static ssize_t iomap_dio_complete(struct iomap_dio *dio)
	74	{
838c4f3d	75	const struct iomap_dio_ops *dops = dio->dops;
db074436 DW	76	struct kiocb *iocb = dio->iocb;
	77	struct inode *inode = file_inode(iocb->ki_filp);
	78	loff_t offset = iocb->ki_pos;
838c4f3d	79	ssize_t ret = dio->error;
db074436	80
838c4f3d CH	81	if (dops && dops->end_io)
838c4f3d CH	82	ret = dops->end_io(iocb, dio->size, ret, dio->flags);
db074436 DW	83
	84	if (likely(!ret)) {
	85	ret = dio->size;
	86	/* check for short read */
	87	if (offset + ret > dio->i_size &&
	88	!(dio->flags & IOMAP_DIO_WRITE))
	89	ret = dio->i_size - offset;
	90	iocb->ki_pos += ret;
	91	}
	92
	93	/*
	94	* Try again to invalidate clean pages which might have been cached by
	95	* non-direct readahead, or faulted in by get_user_pages() if the source
	96	* of the write was an mmap'ed region of the file we're writing. Either
	97	* one is a pretty crazy thing to do, so we don't support it 100%. If
	98	* this invalidation fails, tough, the write still worked...
	99	*
838c4f3d CH	100	* And this page cache invalidation has to be after ->end_io(), as some
	101	* filesystems convert unwritten extents to real allocations in
	102	* ->end_io() when necessary, otherwise a racing buffer read would cache
db074436 DW	103	* zeros from unwritten extents.
	104	*/
	105	if (!dio->error &&
	106	(dio->flags & IOMAP_DIO_WRITE) && inode->i_mapping->nrpages) {
	107	int err;
	108	err = invalidate_inode_pages2_range(inode->i_mapping,
	109	offset >> PAGE_SHIFT,
	110	(offset + dio->size - 1) >> PAGE_SHIFT);
	111	if (err)
	112	dio_warn_stale_pagecache(iocb->ki_filp);
	113	}
	114
	115	/*
	116	* If this is a DSYNC write, make sure we push it to stable storage now
	117	* that we've written data.
	118	*/
	119	if (ret > 0 && (dio->flags & IOMAP_DIO_NEED_SYNC))
	120	ret = generic_write_sync(iocb, ret);
	121
	122	inode_dio_end(file_inode(iocb->ki_filp));
	123	kfree(dio);
	124
	125	return ret;
	126	}
	127
	128	static void iomap_dio_complete_work(struct work_struct *work)
	129	{
	130	struct iomap_dio *dio = container_of(work, struct iomap_dio, aio.work);
	131	struct kiocb *iocb = dio->iocb;
	132
	133	iocb->ki_complete(iocb, iomap_dio_complete(dio), 0);
	134	}
	135
	136	/*
	137	* Set an error in the dio if none is set yet. We have to use cmpxchg
	138	* as the submission context and the completion context(s) can race to
	139	* update the error.
	140	*/
	141	static inline void iomap_dio_set_error(struct iomap_dio *dio, int ret)
	142	{
	143	cmpxchg(&dio->error, 0, ret);
	144	}
	145
	146	static void iomap_dio_bio_end_io(struct bio *bio)
	147	{
	148	struct iomap_dio *dio = bio->bi_private;
	149	bool should_dirty = (dio->flags & IOMAP_DIO_DIRTY);
	150
	151	if (bio->bi_status)
	152	iomap_dio_set_error(dio, blk_status_to_errno(bio->bi_status));
	153
	154	if (atomic_dec_and_test(&dio->ref)) {
	155	if (dio->wait_for_completion) {
	156	struct task_struct *waiter = dio->submit.waiter;
	157	WRITE_ONCE(dio->submit.waiter, NULL);
	158	blk_wake_io_task(waiter);
	159	} else if (dio->flags & IOMAP_DIO_WRITE) {
	160	struct inode *inode = file_inode(dio->iocb->ki_filp);
	161
	162	INIT_WORK(&dio->aio.work, iomap_dio_complete_work);
	163	queue_work(inode->i_sb->s_dio_done_wq, &dio->aio.work);
	164	} else {
	165	iomap_dio_complete_work(&dio->aio.work);
	166	}
167	}
168
169	if (should_dirty) {
170	bio_check_pages_dirty(bio);
171	} else {
172	bio_release_pages(bio, false);
173	bio_put(bio);
174	}
175	}
176
177	static void
178	iomap_dio_zero(struct iomap_dio dio, struct iomap iomap, loff_t pos,
179	unsigned len)
180	{
181	struct page *page = ZERO_PAGE(0);
182	int flags = REQ_SYNC \| REQ_IDLE;
183	struct bio *bio;
184
185	bio = bio_alloc(GFP_KERNEL, 1);
186	bio_set_dev(bio, iomap->bdev);
187	bio->bi_iter.bi_sector = iomap_sector(iomap, pos);
188	bio->bi_private = dio;
189	bio->bi_end_io = iomap_dio_bio_end_io;
190
191	get_page(page);
192	__bio_add_page(bio, page, len, 0);
193	bio_set_op_attrs(bio, REQ_OP_WRITE, flags);
194	iomap_dio_submit_bio(dio, iomap, bio);
195	}
196
197	static loff_t
198	iomap_dio_bio_actor(struct inode *inode, loff_t pos, loff_t length,
199	struct iomap_dio dio, struct iomap iomap)
200	{
201	unsigned int blkbits = blksize_bits(bdev_logical_block_size(iomap->bdev));
202	unsigned int fs_block_size = i_blocksize(inode), pad;
203	unsigned int align = iov_iter_alignment(dio->submit.iter);
204	struct iov_iter iter;
205	struct bio *bio;
206	bool need_zeroout = false;
207	bool use_fua = false;
208	int nr_pages, ret = 0;
209	size_t copied = 0;
210
211	if ((pos \| length \| align) & ((1 << blkbits) - 1))
212	return -EINVAL;
213
214	if (iomap->type == IOMAP_UNWRITTEN) {
215	dio->flags \|= IOMAP_DIO_UNWRITTEN;
216	need_zeroout = true;
217	}
218
219	if (iomap->flags & IOMAP_F_SHARED)
220	dio->flags \|= IOMAP_DIO_COW;
221
222	if (iomap->flags & IOMAP_F_NEW) {
223	need_zeroout = true;
224	} else if (iomap->type == IOMAP_MAPPED) {
225	/*
226	* Use a FUA write if we need datasync semantics, this is a pure
227	* data IO that doesn't require any metadata updates (including
228	* after IO completion such as unwritten extent conversion) and
229	* the underlying device supports FUA. This allows us to avoid
230	* cache flushes on IO completion.
231	*/
232	if (!(iomap->flags & (IOMAP_F_SHARED\|IOMAP_F_DIRTY)) &&
233	(dio->flags & IOMAP_DIO_WRITE_FUA) &&
234	blk_queue_fua(bdev_get_queue(iomap->bdev)))
235	use_fua = true;
236	}
237
238	/*
239	* Operate on a partial iter trimmed to the extent we were called for.
240	* We'll update the iter in the dio once we're done with this extent.
241	*/
242	iter = *dio->submit.iter;
243	iov_iter_truncate(&iter, length);
244
245	nr_pages = iov_iter_npages(&iter, BIO_MAX_PAGES);
246	if (nr_pages <= 0)
247	return nr_pages;
248
249	if (need_zeroout) {
250	/* zero out from the start of the block to the write offset */
251	pad = pos & (fs_block_size - 1);
252	if (pad)
253	iomap_dio_zero(dio, iomap, pos - pad, pad);
254	}
255
256	do {
257	size_t n;
258	if (dio->error) {
259	iov_iter_revert(dio->submit.iter, copied);
260	return 0;
261	}
262
263	bio = bio_alloc(GFP_KERNEL, nr_pages);
264	bio_set_dev(bio, iomap->bdev);
265	bio->bi_iter.bi_sector = iomap_sector(iomap, pos);
266	bio->bi_write_hint = dio->iocb->ki_hint;
267	bio->bi_ioprio = dio->iocb->ki_ioprio;
268	bio->bi_private = dio;
269	bio->bi_end_io = iomap_dio_bio_end_io;
270
271	ret = bio_iov_iter_get_pages(bio, &iter);
272	if (unlikely(ret)) {
273	/*
274	* We have to stop part way through an IO. We must fall
275	* through to the sub-block tail zeroing here, otherwise
276	* this short IO may expose stale data in the tail of
277	* the block we haven't written data to.
278	*/
279	bio_put(bio);
280	goto zero_tail;
281	}
282
283	n = bio->bi_iter.bi_size;
284	if (dio->flags & IOMAP_DIO_WRITE) {
285	bio->bi_opf = REQ_OP_WRITE \| REQ_SYNC \| REQ_IDLE;
286	if (use_fua)
287	bio->bi_opf \|= REQ_FUA;
288	else
289	dio->flags &= ~IOMAP_DIO_WRITE_FUA;
290	task_io_account_write(n);
291	} else {
292	bio->bi_opf = REQ_OP_READ;
293	if (dio->flags & IOMAP_DIO_DIRTY)
294	bio_set_pages_dirty(bio);
295	}
296
297	iov_iter_advance(dio->submit.iter, n);
298
299	dio->size += n;
300	pos += n;
301	copied += n;
302
303	nr_pages = iov_iter_npages(&iter, BIO_MAX_PAGES);
304	iomap_dio_submit_bio(dio, iomap, bio);
305	} while (nr_pages);
306
307	/*
308	* We need to zeroout the tail of a sub-block write if the extent type
309	* requires zeroing or the write extends beyond EOF. If we don't zero
310	* the block tail in the latter case, we can expose stale data via mmap
311	* reads of the EOF block.
312	*/
313	zero_tail:
314	if (need_zeroout \|\|
315	((dio->flags & IOMAP_DIO_WRITE) && pos >= i_size_read(inode))) {
316	/* zero out from the end of the write to the end of the block */
317	pad = pos & (fs_block_size - 1);
318	if (pad)
319	iomap_dio_zero(dio, iomap, pos, fs_block_size - pad);
320	}
321	return copied ? copied : ret;
322	}
323
324	static loff_t
325	iomap_dio_hole_actor(loff_t length, struct iomap_dio *dio)
326	{
327	length = iov_iter_zero(length, dio->submit.iter);
328	dio->size += length;
329	return length;
330	}
331
332	static loff_t
333	iomap_dio_inline_actor(struct inode *inode, loff_t pos, loff_t length,
334	struct iomap_dio dio, struct iomap iomap)
335	{
336	struct iov_iter *iter = dio->submit.iter;
337	size_t copied;
338
339	BUG_ON(pos + length > PAGE_SIZE - offset_in_page(iomap->inline_data));
340
341	if (dio->flags & IOMAP_DIO_WRITE) {
342	loff_t size = inode->i_size;
343
344	if (pos > size)
345	memset(iomap->inline_data + size, 0, pos - size);
346	copied = copy_from_iter(iomap->inline_data + pos, length, iter);
347	if (copied) {
348	if (pos + copied > size)
349	i_size_write(inode, pos + copied);
350	mark_inode_dirty(inode);
351	}
352	} else {
353	copied = copy_to_iter(iomap->inline_data + pos, length, iter);
354	}
355	dio->size += copied;
356	return copied;
357	}
358
359	static loff_t
360	iomap_dio_actor(struct inode *inode, loff_t pos, loff_t length,
c039b997	361	void data, struct iomap iomap, struct iomap *srcmap)
db074436 DW	362	{
	363	struct iomap_dio *dio = data;
	364
	365	switch (iomap->type) {
	366	case IOMAP_HOLE:
	367	if (WARN_ON_ONCE(dio->flags & IOMAP_DIO_WRITE))
	368	return -EIO;
	369	return iomap_dio_hole_actor(length, dio);
	370	case IOMAP_UNWRITTEN:
	371	if (!(dio->flags & IOMAP_DIO_WRITE))
	372	return iomap_dio_hole_actor(length, dio);
	373	return iomap_dio_bio_actor(inode, pos, length, dio, iomap);
	374	case IOMAP_MAPPED:
	375	return iomap_dio_bio_actor(inode, pos, length, dio, iomap);
	376	case IOMAP_INLINE:
	377	return iomap_dio_inline_actor(inode, pos, length, dio, iomap);
	378	default:
	379	WARN_ON_ONCE(1);
	380	return -EIO;
	381	}
	382	}
	383
	384	/*
	385	* iomap_dio_rw() always completes O_[D]SYNC writes regardless of whether the IO
	386	* is being issued as AIO or not. This allows us to optimise pure data writes
	387	* to use REQ_FUA rather than requiring generic_write_sync() to issue a
	388	* REQ_FLUSH post write. This is slightly tricky because a single request here
	389	* can be mapped into multiple disjoint IOs and only a subset of the IOs issued
	390	* may be pure data writes. In that case, we still need to do a full data sync
	391	* completion.
	392	*/
	393	ssize_t
	394	iomap_dio_rw(struct kiocb iocb, struct iov_iter iter,
13ef9544 JK	395	const struct iomap_ops ops, const struct iomap_dio_ops dops,
13ef9544 JK	396	bool wait_for_completion)
db074436 DW	397	{
	398	struct address_space *mapping = iocb->ki_filp->f_mapping;
	399	struct inode *inode = file_inode(iocb->ki_filp);
	400	size_t count = iov_iter_count(iter);
	401	loff_t pos = iocb->ki_pos, start = pos;
	402	loff_t end = iocb->ki_pos + count - 1, ret = 0;
	403	unsigned int flags = IOMAP_DIRECT;
db074436 DW	404	struct blk_plug plug;
	405	struct iomap_dio *dio;
	406
	407	lockdep_assert_held(&inode->i_rwsem);
	408
	409	if (!count)
	410	return 0;
	411
13ef9544 JK	412	if (WARN_ON(is_sync_kiocb(iocb) && !wait_for_completion))
	413	return -EIO;
	414
db074436 DW	415	dio = kmalloc(sizeof(*dio), GFP_KERNEL);
	416	if (!dio)
	417	return -ENOMEM;
	418
	419	dio->iocb = iocb;
	420	atomic_set(&dio->ref, 1);
	421	dio->size = 0;
	422	dio->i_size = i_size_read(inode);
838c4f3d	423	dio->dops = dops;
db074436 DW	424	dio->error = 0;
	425	dio->flags = 0;
	426
	427	dio->submit.iter = iter;
	428	dio->submit.waiter = current;
	429	dio->submit.cookie = BLK_QC_T_NONE;
	430	dio->submit.last_queue = NULL;
	431
	432	if (iov_iter_rw(iter) == READ) {
	433	if (pos >= dio->i_size)
	434	goto out_free_dio;
	435
a9010042	436	if (iter_is_iovec(iter))
db074436 DW	437	dio->flags \|= IOMAP_DIO_DIRTY;
	438	} else {
	439	flags \|= IOMAP_WRITE;
	440	dio->flags \|= IOMAP_DIO_WRITE;
	441
	442	/* for data sync or sync, we need sync completion processing */
	443	if (iocb->ki_flags & IOCB_DSYNC)
	444	dio->flags \|= IOMAP_DIO_NEED_SYNC;
	445
	446	/*
	447	* For datasync only writes, we optimistically try using FUA for
	448	* this IO. Any non-FUA write that occurs will clear this flag,
	449	* hence we know before completion whether a cache flush is
	450	* necessary.
	451	*/
	452	if ((iocb->ki_flags & (IOCB_DSYNC \| IOCB_SYNC)) == IOCB_DSYNC)
	453	dio->flags \|= IOMAP_DIO_WRITE_FUA;
	454	}
	455
	456	if (iocb->ki_flags & IOCB_NOWAIT) {
	457	if (filemap_range_has_page(mapping, start, end)) {
	458	ret = -EAGAIN;
	459	goto out_free_dio;
	460	}
	461	flags \|= IOMAP_NOWAIT;
	462	}
	463
	464	ret = filemap_write_and_wait_range(mapping, start, end);
	465	if (ret)
	466	goto out_free_dio;
	467
	468	/*
	469	* Try to invalidate cache pages for the range we're direct
	470	* writing. If this invalidation fails, tough, the write will
	471	* still work, but racing two incompatible write paths is a
	472	* pretty crazy thing to do, so we don't support it 100%.
	473	*/
	474	ret = invalidate_inode_pages2_range(mapping,
	475	start >> PAGE_SHIFT, end >> PAGE_SHIFT);
	476	if (ret)
	477	dio_warn_stale_pagecache(iocb->ki_filp);
	478	ret = 0;
	479
	480	if (iov_iter_rw(iter) == WRITE && !wait_for_completion &&
	481	!inode->i_sb->s_dio_done_wq) {
	482	ret = sb_init_dio_done_wq(inode->i_sb);
	483	if (ret < 0)
	484	goto out_free_dio;
	485	}
	486
	487	inode_dio_begin(inode);
	488
	489	blk_start_plug(&plug);
	490	do {
	491	ret = iomap_apply(inode, pos, count, flags, ops, dio,
	492	iomap_dio_actor);
	493	if (ret <= 0) {
	494	/* magic error code to fall back to buffered I/O */
	495	if (ret == -ENOTBLK) {
	496	wait_for_completion = true;
	497	ret = 0;
	498	}
	499	break;
	500	}
501	pos += ret;
502
503	if (iov_iter_rw(iter) == READ && pos >= dio->i_size)
504	break;
505	} while ((count = iov_iter_count(iter)) > 0);
506	blk_finish_plug(&plug);
507
508	if (ret < 0)
509	iomap_dio_set_error(dio, ret);
510
511	/*
512	* If all the writes we issued were FUA, we don't need to flush the
513	* cache on IO completion. Clear the sync flag for this case.
514	*/
515	if (dio->flags & IOMAP_DIO_WRITE_FUA)
516	dio->flags &= ~IOMAP_DIO_NEED_SYNC;
517
518	WRITE_ONCE(iocb->ki_cookie, dio->submit.cookie);
519	WRITE_ONCE(iocb->private, dio->submit.last_queue);
520
521	/*
522	* We are about to drop our additional submission reference, which
523	* might be the last reference to the dio. There are three three
524	* different ways we can progress here:
525	*
526	* (a) If this is the last reference we will always complete and free
527	* the dio ourselves.
528	* (b) If this is not the last reference, and we serve an asynchronous
529	* iocb, we must never touch the dio after the decrement, the
530	* I/O completion handler will complete and free it.
531	* (c) If this is not the last reference, but we serve a synchronous
532	* iocb, the I/O completion handler will wake us up on the drop
533	* of the final reference, and we will complete and free it here
534	* after we got woken by the I/O completion handler.
535	*/
536	dio->wait_for_completion = wait_for_completion;
537	if (!atomic_dec_and_test(&dio->ref)) {
538	if (!wait_for_completion)
539	return -EIOCBQUEUED;
540
541	for (;;) {
542	set_current_state(TASK_UNINTERRUPTIBLE);
543	if (!READ_ONCE(dio->submit.waiter))
544	break;
545
546	if (!(iocb->ki_flags & IOCB_HIPRI) \|\|
547	!dio->submit.last_queue \|\|
548	!blk_poll(dio->submit.last_queue,
549	dio->submit.cookie, true))
550	io_schedule();
551	}
552	__set_current_state(TASK_RUNNING);
553	}
554
555	return iomap_dio_complete(dio);
556
557	out_free_dio:
558	kfree(dio);
559	return ret;
560	}
561	EXPORT_SYMBOL_GPL(iomap_dio_rw);