1 files changed, 195 insertions, 213 deletions

diff --git a/fs/direct-io.c b/fs/direct-io.c
index 4a588dbd11b..17e39b047de 100644
--- a/fs/direct-io.c
+++ b/fs/direct-io.c
@@ -37,6 +37,7 @@
 #include <linux/uio.h>
 #include <linux/atomic.h>
 #include <linux/prefetch.h>
+#include <linux/aio.h>
 
 /*
  * How many user pages to map in one call to get_user_pages().  This determines
@@ -70,13 +71,11 @@ struct dio_submit {
 					   been performed at the start of a
 					   write */
 	int pages_in_io;		/* approximate total IO pages */
-	size_t	size;			/* total request size (doesn't change)*/
 	sector_t block_in_file;		/* Current offset into the underlying
 					   file in dio_block units. */
 	unsigned blocks_available;	/* At block_in_file.  changes */
 	int reap_counter;		/* rate limit reaping */
 	sector_t final_block_in_request;/* doesn't change */
-	unsigned first_block_in_page;	/* doesn't change, Used only once */
 	int boundary;			/* prev block is at a boundary */
 	get_block_t *get_block;		/* block mapping function */
 	dio_submit_t *submit_io;	/* IO submition function */
@@ -97,19 +96,14 @@ struct dio_submit {
 	sector_t cur_page_block;	/* Where it starts */
 	loff_t cur_page_fs_offset;	/* Offset in file */
 
-	/*
-	 * Page fetching state. These variables belong to dio_refill_pages().
-	 */
-	int curr_page;			/* changes */
-	int total_pages;		/* doesn't change */
-	unsigned long curr_user_address;/* changes */
-
+	struct iov_iter *iter;
 	/*
 	 * Page queue.  These variables belong to dio_refill_pages() and
 	 * dio_get_page().
 	 */
 	unsigned head;			/* next page to process */
 	unsigned tail;			/* last valid page + 1 */
+	size_t from, to;
 };
 
 /* dio_state communicated between submission path and end_io */
@@ -126,6 +120,7 @@ struct dio {
 	spinlock_t bio_lock;		/* protects BIO fields below */
 	int page_errors;		/* errno from get_user_pages() */
 	int is_async;			/* is IO async ? */
+	bool defer_completion;		/* defer AIO completion to workqueue? */
 	int io_error;			/* IO error in completion path */
 	unsigned long refcount;		/* direct_io_worker() and bios */
 	struct bio *bio_list;		/* singly linked via bi_private */
@@ -140,55 +135,14 @@ struct dio {
 	 * allocation time.  Don't add new fields after pages[] unless you
 	 * wish that they not be zeroed.
 	 */
-	struct page *pages[DIO_PAGES];	/* page buffer */
+	union {
+		struct page *pages[DIO_PAGES];	/* page buffer */
+		struct work_struct complete_work;/* deferred AIO completion */
+	};
 } ____cacheline_aligned_in_smp;
 
 static struct kmem_cache *dio_cache __read_mostly;
 
-static void __inode_dio_wait(struct inode *inode)
-{
-	wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
-	DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
-
-	do {
-		prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
-		if (atomic_read(&inode->i_dio_count))
-			schedule();
-	} while (atomic_read(&inode->i_dio_count));
-	finish_wait(wq, &q.wait);
-}
-
-/**
- * inode_dio_wait - wait for outstanding DIO requests to finish
- * @inode: inode to wait for
- *
- * Waits for all pending direct I/O requests to finish so that we can
- * proceed with a truncate or equivalent operation.
- *
- * Must be called under a lock that serializes taking new references
- * to i_dio_count, usually by inode->i_mutex.
- */
-void inode_dio_wait(struct inode *inode)
-{
-	if (atomic_read(&inode->i_dio_count))
-		__inode_dio_wait(inode);
-}
-EXPORT_SYMBOL_GPL(inode_dio_wait);
-
-/*
- * inode_dio_done - signal finish of a direct I/O requests
- * @inode: inode the direct I/O happens on
- *
- * This is called once we've finished processing a direct I/O request,
- * and is used to wake up callers waiting for direct I/O to be quiesced.
- */
-void inode_dio_done(struct inode *inode)
-{
-	if (atomic_dec_and_test(&inode->i_dio_count))
-		wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);
-}
-EXPORT_SYMBOL_GPL(inode_dio_done);
-
 /*
  * How many pages are in the queue?
  */
@@ -202,15 +156,10 @@ static inline unsigned dio_pages_present(struct dio_submit *sdio)
  */
 static inline int dio_refill_pages(struct dio *dio, struct dio_submit *sdio)
 {
-	int ret;
-	int nr_pages;
+	ssize_t ret;
 
-	nr_pages = min(sdio->total_pages - sdio->curr_page, DIO_PAGES);
-	ret = get_user_pages_fast(
-		sdio->curr_user_address,		/* Where from? */
-		nr_pages,			/* How many pages? */
-		dio->rw == READ,		/* Write to memory? */
-		&dio->pages[0]);		/* Put results here */
+	ret = iov_iter_get_pages(sdio->iter, dio->pages, DIO_PAGES * PAGE_SIZE,
+				&sdio->from);
 
 	if (ret < 0 && sdio->blocks_available && (dio->rw & WRITE)) {
 		struct page *page = ZERO_PAGE(0);
@@ -225,18 +174,19 @@ static inline int dio_refill_pages(struct dio *dio, struct dio_submit *sdio)
 		dio->pages[0] = page;
 		sdio->head = 0;
 		sdio->tail = 1;
-		ret = 0;
-		goto out;
+		sdio->from = 0;
+		sdio->to = PAGE_SIZE;
+		return 0;
 	}
 
 	if (ret >= 0) {
-		sdio->curr_user_address += ret * PAGE_SIZE;
-		sdio->curr_page += ret;
+		iov_iter_advance(sdio->iter, ret);
+		ret += sdio->from;
 		sdio->head = 0;
-		sdio->tail = ret;
-		ret = 0;
+		sdio->tail = (ret + PAGE_SIZE - 1) / PAGE_SIZE;
+		sdio->to = ((ret - 1) & (PAGE_SIZE - 1)) + 1;
+		return 0;
 	}
-out:
 	return ret;	
 }
 
@@ -247,7 +197,7 @@ out:
  * L1 cache.
  */
 static inline struct page *dio_get_page(struct dio *dio,
-		struct dio_submit *sdio)
+					struct dio_submit *sdio)
 {
 	if (dio_pages_present(sdio) == 0) {
 		int ret;
@@ -257,23 +207,23 @@ static inline struct page *dio_get_page(struct dio *dio,
 			return ERR_PTR(ret);
 		BUG_ON(dio_pages_present(sdio) == 0);
 	}
-	return dio->pages[sdio->head++];
+	return dio->pages[sdio->head];
 }
 
 /**
  * dio_complete() - called when all DIO BIO I/O has been completed
  * @offset: the byte offset in the file of the completed operation
  *
- * This releases locks as dictated by the locking type, lets interested parties
- * know that a DIO operation has completed, and calculates the resulting return
- * code for the operation.
+ * This drops i_dio_count, lets interested parties know that a DIO operation
+ * has completed, and calculates the resulting return code for the operation.
  *
  * It lets the filesystem know if it registered an interest earlier via
  * get_block.  Pass the private field of the map buffer_head so that
  * filesystems can use it to hold additional state between get_block calls and
  * dio_complete.
  */
-static ssize_t dio_complete(struct dio *dio, loff_t offset, ssize_t ret, bool is_async)
+static ssize_t dio_complete(struct dio *dio, loff_t offset, ssize_t ret,
+		bool is_async)
 {
 	ssize_t transferred = 0;
 
@@ -301,19 +251,36 @@ static ssize_t dio_complete(struct dio *dio, loff_t offset, ssize_t ret, bool is
 	if (ret == 0)
 		ret = transferred;
 
-	if (dio->end_io && dio->result) {
-		dio->end_io(dio->iocb, offset, transferred,
-			    dio->private, ret, is_async);
-	} else {
-		if (is_async)
-			aio_complete(dio->iocb, ret, 0);
-		inode_dio_done(dio->inode);
+	if (dio->end_io && dio->result)
+		dio->end_io(dio->iocb, offset, transferred, dio->private);
+
+	inode_dio_done(dio->inode);
+	if (is_async) {
+		if (dio->rw & WRITE) {
+			int err;
+
+			err = generic_write_sync(dio->iocb->ki_filp, offset,
+						 transferred);
+			if (err < 0 && ret > 0)
+				ret = err;
+		}
+
+		aio_complete(dio->iocb, ret, 0);
 	}
 
+	kmem_cache_free(dio_cache, dio);
 	return ret;
 }
 
+static void dio_aio_complete_work(struct work_struct *work)
+{
+	struct dio *dio = container_of(work, struct dio, complete_work);
+
+	dio_complete(dio, dio->iocb->ki_pos, 0, true);
+}
+
 static int dio_bio_complete(struct dio *dio, struct bio *bio);
+
 /*
  * Asynchronous IO callback. 
  */
@@ -333,8 +300,13 @@ static void dio_bio_end_aio(struct bio *bio, int error)
 	spin_unlock_irqrestore(&dio->bio_lock, flags);
 
 	if (remaining == 0) {
-		dio_complete(dio, dio->iocb->ki_pos, 0, true);
-		kmem_cache_free(dio_cache, dio);
+		if (dio->result && dio->defer_completion) {
+			INIT_WORK(&dio->complete_work, dio_aio_complete_work);
+			queue_work(dio->inode->i_sb->s_dio_done_wq,
+				   &dio->complete_work);
+		} else {
+			dio_complete(dio, dio->iocb->ki_pos, 0, true);
+		}
 	}
 }
 
@@ -392,7 +364,7 @@ dio_bio_alloc(struct dio *dio, struct dio_submit *sdio,
 	bio = bio_alloc(GFP_KERNEL, nr_vecs);
 
 	bio->bi_bdev = bdev;
-	bio->bi_sector = first_sector;
+	bio->bi_iter.bi_sector = first_sector;
 	if (dio->is_async)
 		bio->bi_end_io = dio_bio_end_aio;
 	else
@@ -439,8 +411,8 @@ static inline void dio_bio_submit(struct dio *dio, struct dio_submit *sdio)
  */
 static inline void dio_cleanup(struct dio *dio, struct dio_submit *sdio)
 {
-	while (dio_pages_present(sdio))
-		page_cache_release(dio_get_page(dio, sdio));
+	while (sdio->head < sdio->tail)
+		page_cache_release(dio->pages[sdio->head++]);
 }
 
 /*
@@ -485,8 +457,8 @@ static struct bio *dio_await_one(struct dio *dio)
 static int dio_bio_complete(struct dio *dio, struct bio *bio)
 {
 	const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
-	struct bio_vec *bvec = bio->bi_io_vec;
-	int page_no;
+	struct bio_vec *bvec;
+	unsigned i;
 
 	if (!uptodate)
 		dio->io_error = -EIO;
@@ -494,8 +466,8 @@ static int dio_bio_complete(struct dio *dio, struct bio *bio)
 	if (dio->is_async && dio->rw == READ) {
 		bio_check_pages_dirty(bio);	/* transfers ownership */
 	} else {
-		for (page_no = 0; page_no < bio->bi_vcnt; page_no++) {
-			struct page *page = bvec[page_no].bv_page;
+		bio_for_each_segment_all(bvec, bio, i) {
+			struct page *page = bvec->bv_page;
 
 			if (dio->rw == READ && !PageCompound(page))
 				set_page_dirty_lock(page);
@@ -554,6 +526,42 @@ static inline int dio_bio_reap(struct dio *dio, struct dio_submit *sdio)
 }
 
 /*
+ * Create workqueue for deferred direct IO completions. We allocate the
+ * workqueue when it's first needed. This avoids creating workqueue for
+ * filesystems that don't need it and also allows us to create the workqueue
+ * late enough so the we can include s_id in the name of the workqueue.
+ */
+static int sb_init_dio_done_wq(struct super_block *sb)
+{
+	struct workqueue_struct *old;
+	struct workqueue_struct *wq = alloc_workqueue("dio/%s",
+						      WQ_MEM_RECLAIM, 0,
+						      sb->s_id);
+	if (!wq)
+		return -ENOMEM;
+	/*
+	 * This has to be atomic as more DIOs can race to create the workqueue
+	 */
+	old = cmpxchg(&sb->s_dio_done_wq, NULL, wq);
+	/* Someone created workqueue before us? Free ours... */
+	if (old)
+		destroy_workqueue(wq);
+	return 0;
+}
+
+static int dio_set_defer_completion(struct dio *dio)
+{
+	struct super_block *sb = dio->inode->i_sb;
+
+	if (dio->defer_completion)
+		return 0;
+	dio->defer_completion = true;
+	if (!sb->s_dio_done_wq)
+		return sb_init_dio_done_wq(sb);
+	return 0;
+}
+
+/*
  * Call into the fs to map some more disk blocks.  We record the current number
  * of available blocks at sdio->blocks_available.  These are in units of the
  * fs blocksize, (1 << inode->i_blkbits).
@@ -584,6 +592,7 @@ static int get_more_blocks(struct dio *dio, struct dio_submit *sdio,
 	sector_t fs_endblk;	/* Into file, in filesystem-sized blocks */
 	unsigned long fs_count;	/* Number of filesystem-sized blocks */
 	int create;
+	unsigned int i_blkbits = sdio->blkbits + sdio->blkfactor;
 
 	/*
 	 * If there was a memory error and we've overwritten all the
@@ -598,7 +607,7 @@ static int get_more_blocks(struct dio *dio, struct dio_submit *sdio,
 		fs_count = fs_endblk - fs_startblk + 1;
 
 		map_bh->b_state = 0;
-		map_bh->b_size = fs_count << dio->inode->i_blkbits;
+		map_bh->b_size = fs_count << i_blkbits;
 
 		/*
 		 * For writes inside i_size on a DIO_SKIP_HOLES filesystem we
@@ -623,6 +632,9 @@ static int get_more_blocks(struct dio *dio, struct dio_submit *sdio,
 
 		/* Store for completion */
 		dio->private = map_bh->b_private;
+
+		if (ret == 0 && buffer_defer_completion(map_bh))
+			ret = dio_set_defer_completion(dio);
 	}
 	return ret;
 }
@@ -641,7 +653,6 @@ static inline int dio_new_bio(struct dio *dio, struct dio_submit *sdio,
 		goto out;
 	sector = start_sector << (sdio->blkbits - 9);
 	nr_pages = min(sdio->pages_in_io, bio_get_nr_vecs(map_bh->b_bdev));
-	nr_pages = min(nr_pages, BIO_MAX_PAGES);
 	BUG_ON(nr_pages <= 0);
 	dio_bio_alloc(dio, sdio, map_bh->b_bdev, sector, nr_pages);
 	sdio->boundary = 0;
@@ -696,7 +707,7 @@ static inline int dio_send_cur_page(struct dio *dio, struct dio_submit *sdio,
 	if (sdio->bio) {
 		loff_t cur_offset = sdio->cur_page_fs_offset;
 		loff_t bio_next_offset = sdio->logical_offset_in_bio +
-			sdio->bio->bi_size;
+			sdio->bio->bi_iter.bi_size;
 
 		/*
 		 * See whether this new request is contiguous with the old.
@@ -715,12 +726,6 @@ static inline int dio_send_cur_page(struct dio *dio, struct dio_submit *sdio,
 		if (sdio->final_block_in_bio != sdio->cur_page_block ||
 		    cur_offset != bio_next_offset)
 			dio_bio_submit(dio, sdio);
-		/*
-		 * Submit now if the underlying fs is about to perform a
-		 * metadata read
-		 */
-		else if (sdio->boundary)
-			dio_bio_submit(dio, sdio);
 	}
 
 	if (sdio->bio == NULL) {
@@ -780,16 +785,6 @@ submit_page_section(struct dio *dio, struct dio_submit *sdio, struct page *page,
 	    sdio->cur_page_block +
 	    (sdio->cur_page_len >> sdio->blkbits) == blocknr) {
 		sdio->cur_page_len += len;
-
-		/*
-		 * If sdio->boundary then we want to schedule the IO now to
-		 * avoid metadata seeks.
-		 */
-		if (sdio->boundary) {
-			ret = dio_send_cur_page(dio, sdio, map_bh);
-			page_cache_release(sdio->cur_page);
-			sdio->cur_page = NULL;
-		}
 		goto out;
 	}
 
@@ -801,7 +796,7 @@ submit_page_section(struct dio *dio, struct dio_submit *sdio, struct page *page,
 		page_cache_release(sdio->cur_page);
 		sdio->cur_page = NULL;
 		if (ret)
-			goto out;
+			return ret;
 	}
 
 	page_cache_get(page);		/* It is in dio */
@@ -811,6 +806,16 @@ submit_page_section(struct dio *dio, struct dio_submit *sdio, struct page *page,
 	sdio->cur_page_block = blocknr;
 	sdio->cur_page_fs_offset = sdio->block_in_file << sdio->blkbits;
 out:
+	/*
+	 * If sdio->boundary then we want to schedule the IO now to
+	 * avoid metadata seeks.
+	 */
+	if (sdio->boundary) {
+		ret = dio_send_cur_page(dio, sdio, map_bh);
+		dio_bio_submit(dio, sdio);
+		page_cache_release(sdio->cur_page);
+		sdio->cur_page = NULL;
+	}
 	return ret;
 }
 
@@ -896,23 +901,22 @@ static int do_direct_IO(struct dio *dio, struct dio_submit *sdio,
 			struct buffer_head *map_bh)
 {
 	const unsigned blkbits = sdio->blkbits;
-	const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
-	struct page *page;
-	unsigned block_in_page;
 	int ret = 0;
 
-	/* The I/O can start at any block offset within the first page */
-	block_in_page = sdio->first_block_in_page;
-
 	while (sdio->block_in_file < sdio->final_block_in_request) {
+		struct page *page;
+		size_t from, to;
+
 		page = dio_get_page(dio, sdio);
 		if (IS_ERR(page)) {
 			ret = PTR_ERR(page);
 			goto out;
 		}
+		from = sdio->head ? 0 : sdio->from;
+		to = (sdio->head == sdio->tail - 1) ? sdio->to : PAGE_SIZE;
+		sdio->head++;
 
-		while (block_in_page < blocks_per_page) {
-			unsigned offset_in_page = block_in_page << blkbits;
+		while (from < to) {
 			unsigned this_chunk_bytes;	/* # of bytes mapped */
 			unsigned this_chunk_blocks;	/* # of blocks */
 			unsigned u;
@@ -983,10 +987,10 @@ do_holes:
 					page_cache_release(page);
 					goto out;
 				}
-				zero_user(page, block_in_page << blkbits,
-						1 << blkbits);
+				zero_user(page, from, 1 << blkbits);
 				sdio->block_in_file++;
-				block_in_page++;
+				from += 1 << blkbits;
+				dio->result += 1 << blkbits;
 				goto next_block;
 			}
 
@@ -1003,7 +1007,7 @@ do_holes:
 			 * can add to this page
 			 */
 			this_chunk_blocks = sdio->blocks_available;
-			u = (PAGE_SIZE - offset_in_page) >> blkbits;
+			u = (to - from) >> blkbits;
 			if (this_chunk_blocks > u)
 				this_chunk_blocks = u;
 			u = sdio->final_block_in_request - sdio->block_in_file;
@@ -1012,9 +1016,10 @@ do_holes:
 			this_chunk_bytes = this_chunk_blocks << blkbits;
 			BUG_ON(this_chunk_bytes == 0);
 
-			sdio->boundary = buffer_boundary(map_bh);
+			if (this_chunk_blocks == sdio->blocks_available)
+				sdio->boundary = buffer_boundary(map_bh);
 			ret = submit_page_section(dio, sdio, page,
-						  offset_in_page,
+						  from,
 						  this_chunk_bytes,
 						  sdio->next_block_for_io,
 						  map_bh);
@@ -1025,7 +1030,8 @@ do_holes:
 			sdio->next_block_for_io += this_chunk_blocks;
 
 			sdio->block_in_file += this_chunk_blocks;
-			block_in_page += this_chunk_blocks;
+			from += this_chunk_bytes;
+			dio->result += this_chunk_bytes;
 			sdio->blocks_available -= this_chunk_blocks;
 next_block:
 			BUG_ON(sdio->block_in_file > sdio->final_block_in_request);
@@ -1035,7 +1041,6 @@ next_block:
 
 		/* Drop the ref which was taken in get_user_pages() */
 		page_cache_release(page);
-		block_in_page = 0;
 	}
 out:
 	return ret;
@@ -1090,22 +1095,21 @@ static inline int drop_refcount(struct dio *dio)
  */
 static inline ssize_t
 do_blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
-	struct block_device *bdev, const struct iovec *iov, loff_t offset, 
-	unsigned long nr_segs, get_block_t get_block, dio_iodone_t end_io,
+	struct block_device *bdev, struct iov_iter *iter, loff_t offset, 
+	get_block_t get_block, dio_iodone_t end_io,
 	dio_submit_t submit_io,	int flags)
 {
-	int seg;
-	size_t size;
-	unsigned long addr;
-	unsigned blkbits = inode->i_blkbits;
+	unsigned i_blkbits = ACCESS_ONCE(inode->i_blkbits);
+	unsigned blkbits = i_blkbits;
 	unsigned blocksize_mask = (1 << blkbits) - 1;
 	ssize_t retval = -EINVAL;
-	loff_t end = offset;
+	size_t count = iov_iter_count(iter);
+	loff_t end = offset + count;
 	struct dio *dio;
 	struct dio_submit sdio = { 0, };
-	unsigned long user_addr;
-	size_t bytes;
 	struct buffer_head map_bh = { 0, };
+	struct blk_plug plug;
+	unsigned long align = offset | iov_iter_alignment(iter);
 
 	if (rw & WRITE)
 		rw = WRITE_ODIRECT;
@@ -1115,32 +1119,16 @@ do_blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
 	 * the early prefetch in the caller enough time.
 	 */
 
-	if (offset & blocksize_mask) {
+	if (align & blocksize_mask) {
 		if (bdev)
 			blkbits = blksize_bits(bdev_logical_block_size(bdev));
 		blocksize_mask = (1 << blkbits) - 1;
-		if (offset & blocksize_mask)
+		if (align & blocksize_mask)
 			goto out;
 	}
 
-	/* Check the memory alignment.  Blocks cannot straddle pages */
-	for (seg = 0; seg < nr_segs; seg++) {
-		addr = (unsigned long)iov[seg].iov_base;
-		size = iov[seg].iov_len;
-		end += size;
-		if (unlikely((addr & blocksize_mask) ||
-			     (size & blocksize_mask))) {
-			if (bdev)
-				blkbits = blksize_bits(
-					 bdev_logical_block_size(bdev));
-			blocksize_mask = (1 << blkbits) - 1;
-			if ((addr & blocksize_mask) || (size & blocksize_mask))
-				goto out;
-		}
-	}
-
 	/* watch out for a 0 len io from a tricksy fs */
-	if (rw == READ && end == offset)
+	if (rw == READ && !iov_iter_count(iter))
 		return 0;
 
 	dio = kmem_cache_alloc(dio_cache, GFP_KERNEL);
@@ -1174,25 +1162,48 @@ do_blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
 	}
 
 	/*
-	 * Will be decremented at I/O completion time.
+	 * For file extending writes updating i_size before data writeouts
+	 * complete can expose uninitialized blocks in dumb filesystems.
+	 * In that case we need to wait for I/O completion even if asked
+	 * for an asynchronous write.
 	 */
-	atomic_inc(&inode->i_dio_count);
+	if (is_sync_kiocb(iocb))
+		dio->is_async = false;
+	else if (!(dio->flags & DIO_ASYNC_EXTEND) &&
+            (rw & WRITE) && end > i_size_read(inode))
+		dio->is_async = false;
+	else
+		dio->is_async = true;
+
+	dio->inode = inode;
+	dio->rw = rw;
 
 	/*
-	 * For file extending writes updating i_size before data
-	 * writeouts complete can expose uninitialized blocks. So
-	 * even for AIO, we need to wait for i/o to complete before
-	 * returning in this case.
+	 * For AIO O_(D)SYNC writes we need to defer completions to a workqueue
+	 * so that we can call ->fsync.
 	 */
-	dio->is_async = !is_sync_kiocb(iocb) && !((rw & WRITE) &&
-		(end > i_size_read(inode)));
+	if (dio->is_async && (rw & WRITE) &&
+	    ((iocb->ki_filp->f_flags & O_DSYNC) ||
+	     IS_SYNC(iocb->ki_filp->f_mapping->host))) {
+		retval = dio_set_defer_completion(dio);
+		if (retval) {
+			/*
+			 * We grab i_mutex only for reads so we don't have
+			 * to release it here
+			 */
+			kmem_cache_free(dio_cache, dio);
+			goto out;
+		}
+	}
 
-	retval = 0;
+	/*
+	 * Will be decremented at I/O completion time.
+	 */
+	atomic_inc(&inode->i_dio_count);
 
-	dio->inode = inode;
-	dio->rw = rw;
+	retval = 0;
 	sdio.blkbits = blkbits;
-	sdio.blkfactor = inode->i_blkbits - blkbits;
+	sdio.blkfactor = i_blkbits - blkbits;
 	sdio.block_in_file = offset >> blkbits;
 
 	sdio.get_block = get_block;
@@ -1207,6 +1218,10 @@ do_blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
 	spin_lock_init(&dio->bio_lock);
 	dio->refcount = 1;
 
+	sdio.iter = iter;
+	sdio.final_block_in_request =
+		(offset + iov_iter_count(iter)) >> blkbits;
+
 	/*
 	 * In case of non-aligned buffers, we may need 2 more
 	 * pages since we need to zero out first and last block.
@@ -1214,45 +1229,13 @@ do_blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
 	if (unlikely(sdio.blkfactor))
 		sdio.pages_in_io = 2;
 
-	for (seg = 0; seg < nr_segs; seg++) {
-		user_addr = (unsigned long)iov[seg].iov_base;
-		sdio.pages_in_io +=
-			((user_addr + iov[seg].iov_len + PAGE_SIZE-1) /
-				PAGE_SIZE - user_addr / PAGE_SIZE);
-	}
-
-	for (seg = 0; seg < nr_segs; seg++) {
-		user_addr = (unsigned long)iov[seg].iov_base;
-		sdio.size += bytes = iov[seg].iov_len;
-
-		/* Index into the first page of the first block */
-		sdio.first_block_in_page = (user_addr & ~PAGE_MASK) >> blkbits;
-		sdio.final_block_in_request = sdio.block_in_file +
-						(bytes >> blkbits);
-		/* Page fetching state */
-		sdio.head = 0;
-		sdio.tail = 0;
-		sdio.curr_page = 0;
-
-		sdio.total_pages = 0;
-		if (user_addr & (PAGE_SIZE-1)) {
-			sdio.total_pages++;
-			bytes -= PAGE_SIZE - (user_addr & (PAGE_SIZE - 1));
-		}
-		sdio.total_pages += (bytes + PAGE_SIZE - 1) / PAGE_SIZE;
-		sdio.curr_user_address = user_addr;
+	sdio.pages_in_io += iov_iter_npages(iter, INT_MAX);
 
-		retval = do_direct_IO(dio, &sdio, &map_bh);
+	blk_start_plug(&plug);
 
-		dio->result += iov[seg].iov_len -
-			((sdio.final_block_in_request - sdio.block_in_file) <<
-					blkbits);
-
-		if (retval) {
-			dio_cleanup(dio, &sdio);
-			break;
-		}
-	} /* end iovec loop */
+	retval = do_direct_IO(dio, &sdio, &map_bh);
+	if (retval)
+		dio_cleanup(dio, &sdio);
 
 	if (retval == -ENOTBLK) {
 		/*
@@ -1279,6 +1262,8 @@ do_blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
 	if (sdio.bio)
 		dio_bio_submit(dio, &sdio);
 
+	blk_finish_plug(&plug);
+
 	/*
 	 * It is possible that, we return short IO due to end of file.
 	 * In that case, we need to release all the pages we got hold on.
@@ -1302,15 +1287,13 @@ do_blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
 	 */
 	BUG_ON(retval == -EIOCBQUEUED);
 	if (dio->is_async && retval == 0 && dio->result &&
-	    ((rw & READ) || (dio->result == sdio.size)))
+	    (rw == READ || dio->result == count))
 		retval = -EIOCBQUEUED;
-
-	if (retval != -EIOCBQUEUED)
+	else
 		dio_await_completion(dio);
 
 	if (drop_refcount(dio) == 0) {
 		retval = dio_complete(dio, offset, retval, false);
-		kmem_cache_free(dio_cache, dio);
 	} else
 		BUG_ON(retval != -EIOCBQUEUED);
 
@@ -1320,8 +1303,8 @@ out:
 
 ssize_t
 __blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
-	struct block_device *bdev, const struct iovec *iov, loff_t offset,
-	unsigned long nr_segs, get_block_t get_block, dio_iodone_t end_io,
+	struct block_device *bdev, struct iov_iter *iter, loff_t offset,
+	get_block_t get_block, dio_iodone_t end_io,
 	dio_submit_t submit_io,	int flags)
 {
 	/*
@@ -1336,9 +1319,8 @@ __blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
 	prefetch(bdev->bd_queue);
 	prefetch((char *)bdev->bd_queue + SMP_CACHE_BYTES);
 
-	return do_blockdev_direct_IO(rw, iocb, inode, bdev, iov, offset,
-				     nr_segs, get_block, end_io,
-				     submit_io, flags);
+	return do_blockdev_direct_IO(rw, iocb, inode, bdev, iter, offset,
+				     get_block, end_io, submit_io, flags);
 }
 
 EXPORT_SYMBOL(__blockdev_direct_IO);