1 files changed, 313 insertions, 247 deletions

diff --git a/fs/ext4/page-io.c b/fs/ext4/page-io.c
index 7f5451cd1d3..b24a2541a9b 100644
--- a/fs/ext4/page-io.c
+++ b/fs/ext4/page-io.c
@@ -6,7 +6,6 @@
  * Written by Theodore Ts'o, 2010.
  */
 
-#include <linux/module.h>
 #include <linux/fs.h>
 #include <linux/time.h>
 #include <linux/jbd2.h>
@@ -19,258 +18,323 @@
 #include <linux/pagevec.h>
 #include <linux/mpage.h>
 #include <linux/namei.h>
+#include <linux/aio.h>
 #include <linux/uio.h>
 #include <linux/bio.h>
 #include <linux/workqueue.h>
 #include <linux/kernel.h>
 #include <linux/slab.h>
+#include <linux/mm.h>
+#include <linux/ratelimit.h>
 
 #include "ext4_jbd2.h"
 #include "xattr.h"
 #include "acl.h"
-#include "ext4_extents.h"
 
-static struct kmem_cache *io_page_cachep, *io_end_cachep;
-
-#define WQ_HASH_SZ		37
-#define to_ioend_wq(v)	(&ioend_wq[((unsigned long)v) % WQ_HASH_SZ])
-static wait_queue_head_t ioend_wq[WQ_HASH_SZ];
+static struct kmem_cache *io_end_cachep;
 
 int __init ext4_init_pageio(void)
 {
-	int i;
-
-	io_page_cachep = KMEM_CACHE(ext4_io_page, SLAB_RECLAIM_ACCOUNT);
-	if (io_page_cachep == NULL)
-		return -ENOMEM;
 	io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
-	if (io_page_cachep == NULL) {
-		kmem_cache_destroy(io_page_cachep);
+	if (io_end_cachep == NULL)
 		return -ENOMEM;
-	}
-	for (i = 0; i < WQ_HASH_SZ; i++)
-		init_waitqueue_head(&ioend_wq[i]);
-
 	return 0;
 }
 
 void ext4_exit_pageio(void)
 {
 	kmem_cache_destroy(io_end_cachep);
-	kmem_cache_destroy(io_page_cachep);
 }
 
-void ext4_ioend_wait(struct inode *inode)
+/*
+ * Print an buffer I/O error compatible with the fs/buffer.c.  This
+ * provides compatibility with dmesg scrapers that look for a specific
+ * buffer I/O error message.  We really need a unified error reporting
+ * structure to userspace ala Digital Unix's uerf system, but it's
+ * probably not going to happen in my lifetime, due to LKML politics...
+ */
+static void buffer_io_error(struct buffer_head *bh)
+{
+	char b[BDEVNAME_SIZE];
+	printk_ratelimited(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n",
+			bdevname(bh->b_bdev, b),
+			(unsigned long long)bh->b_blocknr);
+}
+
+static void ext4_finish_bio(struct bio *bio)
 {
-	wait_queue_head_t *wq = to_ioend_wq(inode);
+	int i;
+	int error = !test_bit(BIO_UPTODATE, &bio->bi_flags);
+	struct bio_vec *bvec;
+
+	bio_for_each_segment_all(bvec, bio, i) {
+		struct page *page = bvec->bv_page;
+		struct buffer_head *bh, *head;
+		unsigned bio_start = bvec->bv_offset;
+		unsigned bio_end = bio_start + bvec->bv_len;
+		unsigned under_io = 0;
+		unsigned long flags;
 
-	wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_ioend_count) == 0));
+		if (!page)
+			continue;
+
+		if (error) {
+			SetPageError(page);
+			set_bit(AS_EIO, &page->mapping->flags);
+		}
+		bh = head = page_buffers(page);
+		/*
+		 * We check all buffers in the page under BH_Uptodate_Lock
+		 * to avoid races with other end io clearing async_write flags
+		 */
+		local_irq_save(flags);
+		bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
+		do {
+			if (bh_offset(bh) < bio_start ||
+			    bh_offset(bh) + bh->b_size > bio_end) {
+				if (buffer_async_write(bh))
+					under_io++;
+				continue;
+			}
+			clear_buffer_async_write(bh);
+			if (error)
+				buffer_io_error(bh);
+		} while ((bh = bh->b_this_page) != head);
+		bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
+		local_irq_restore(flags);
+		if (!under_io)
+			end_page_writeback(page);
+	}
 }
 
-static void put_io_page(struct ext4_io_page *io_page)
+static void ext4_release_io_end(ext4_io_end_t *io_end)
 {
-	if (atomic_dec_and_test(&io_page->p_count)) {
-		end_page_writeback(io_page->p_page);
-		put_page(io_page->p_page);
-		kmem_cache_free(io_page_cachep, io_page);
+	struct bio *bio, *next_bio;
+
+	BUG_ON(!list_empty(&io_end->list));
+	BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
+	WARN_ON(io_end->handle);
+
+	if (atomic_dec_and_test(&EXT4_I(io_end->inode)->i_ioend_count))
+		wake_up_all(ext4_ioend_wq(io_end->inode));
+
+	for (bio = io_end->bio; bio; bio = next_bio) {
+		next_bio = bio->bi_private;
+		ext4_finish_bio(bio);
+		bio_put(bio);
 	}
+	kmem_cache_free(io_end_cachep, io_end);
 }
 
-void ext4_free_io_end(ext4_io_end_t *io)
+static void ext4_clear_io_unwritten_flag(ext4_io_end_t *io_end)
 {
-	int i;
-	wait_queue_head_t *wq;
-
-	BUG_ON(!io);
-	if (io->page)
-		put_page(io->page);
-	for (i = 0; i < io->num_io_pages; i++)
-		put_io_page(io->pages[i]);
-	io->num_io_pages = 0;
-	wq = to_ioend_wq(io->inode);
-	if (atomic_dec_and_test(&EXT4_I(io->inode)->i_ioend_count) &&
-	    waitqueue_active(wq))
-		wake_up_all(wq);
-	kmem_cache_free(io_end_cachep, io);
+	struct inode *inode = io_end->inode;
+
+	io_end->flag &= ~EXT4_IO_END_UNWRITTEN;
+	/* Wake up anyone waiting on unwritten extent conversion */
+	if (atomic_dec_and_test(&EXT4_I(inode)->i_unwritten))
+		wake_up_all(ext4_ioend_wq(inode));
 }
 
 /*
- * check a range of space and convert unwritten extents to written.
+ * Check a range of space and convert unwritten extents to written. Note that
+ * we are protected from truncate touching same part of extent tree by the
+ * fact that truncate code waits for all DIO to finish (thus exclusion from
+ * direct IO is achieved) and also waits for PageWriteback bits. Thus we
+ * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
+ * completed (happens from ext4_free_ioend()).
  */
-int ext4_end_io_nolock(ext4_io_end_t *io)
+static int ext4_end_io(ext4_io_end_t *io)
 {
 	struct inode *inode = io->inode;
 	loff_t offset = io->offset;
 	ssize_t size = io->size;
+	handle_t *handle = io->handle;
 	int ret = 0;
 
 	ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
 		   "list->prev 0x%p\n",
 		   io, inode->i_ino, io->list.next, io->list.prev);
 
-	if (list_empty(&io->list))
-		return ret;
-
-	if (!(io->flag & EXT4_IO_END_UNWRITTEN))
-		return ret;
-
-	ret = ext4_convert_unwritten_extents(inode, offset, size);
+	io->handle = NULL;	/* Following call will use up the handle */
+	ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
 	if (ret < 0) {
-		printk(KERN_EMERG "%s: failed to convert unwritten "
-			"extents to written extents, error is %d "
-			"io is still on inode %lu aio dio list\n",
-		       __func__, ret, inode->i_ino);
-		return ret;
+		ext4_msg(inode->i_sb, KERN_EMERG,
+			 "failed to convert unwritten extents to written "
+			 "extents -- potential data loss!  "
+			 "(inode %lu, offset %llu, size %zd, error %d)",
+			 inode->i_ino, offset, size, ret);
 	}
-
-	if (io->iocb)
-		aio_complete(io->iocb, io->result, 0);
-	/* clear the DIO AIO unwritten flag */
-	io->flag &= ~EXT4_IO_END_UNWRITTEN;
+	ext4_clear_io_unwritten_flag(io);
+	ext4_release_io_end(io);
 	return ret;
 }
 
-/*
- * work on completed aio dio IO, to convert unwritten extents to extents
- */
-static void ext4_end_io_work(struct work_struct *work)
+static void dump_completed_IO(struct inode *inode, struct list_head *head)
 {
-	ext4_io_end_t		*io = container_of(work, ext4_io_end_t, work);
-	struct inode		*inode = io->inode;
-	struct ext4_inode_info	*ei = EXT4_I(inode);
-	unsigned long		flags;
-	int			ret;
-
-	mutex_lock(&inode->i_mutex);
-	ret = ext4_end_io_nolock(io);
-	if (ret < 0) {
-		mutex_unlock(&inode->i_mutex);
+#ifdef	EXT4FS_DEBUG
+	struct list_head *cur, *before, *after;
+	ext4_io_end_t *io, *io0, *io1;
+
+	if (list_empty(head))
 		return;
+
+	ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
+	list_for_each_entry(io, head, list) {
+		cur = &io->list;
+		before = cur->prev;
+		io0 = container_of(before, ext4_io_end_t, list);
+		after = cur->next;
+		io1 = container_of(after, ext4_io_end_t, list);
+
+		ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
+			    io, inode->i_ino, io0, io1);
 	}
+#endif
+}
 
+/* Add the io_end to per-inode completed end_io list. */
+static void ext4_add_complete_io(ext4_io_end_t *io_end)
+{
+	struct ext4_inode_info *ei = EXT4_I(io_end->inode);
+	struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
+	struct workqueue_struct *wq;
+	unsigned long flags;
+
+	/* Only reserved conversions from writeback should enter here */
+	WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
+	WARN_ON(!io_end->handle && sbi->s_journal);
 	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
-	if (!list_empty(&io->list))
-		list_del_init(&io->list);
+	wq = sbi->rsv_conversion_wq;
+	if (list_empty(&ei->i_rsv_conversion_list))
+		queue_work(wq, &ei->i_rsv_conversion_work);
+	list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
 	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
-	mutex_unlock(&inode->i_mutex);
-	ext4_free_io_end(io);
 }
 
-ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
+static int ext4_do_flush_completed_IO(struct inode *inode,
+				      struct list_head *head)
 {
-	ext4_io_end_t *io = NULL;
+	ext4_io_end_t *io;
+	struct list_head unwritten;
+	unsigned long flags;
+	struct ext4_inode_info *ei = EXT4_I(inode);
+	int err, ret = 0;
 
-	io = kmem_cache_alloc(io_end_cachep, flags);
+	spin_lock_irqsave(&ei->i_completed_io_lock, flags);
+	dump_completed_IO(inode, head);
+	list_replace_init(head, &unwritten);
+	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
+
+	while (!list_empty(&unwritten)) {
+		io = list_entry(unwritten.next, ext4_io_end_t, list);
+		BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
+		list_del_init(&io->list);
+
+		err = ext4_end_io(io);
+		if (unlikely(!ret && err))
+			ret = err;
+	}
+	return ret;
+}
+
+/*
+ * work on completed IO, to convert unwritten extents to extents
+ */
+void ext4_end_io_rsv_work(struct work_struct *work)
+{
+	struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
+						  i_rsv_conversion_work);
+	ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
+}
+
+ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
+{
+	ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
 	if (io) {
-		memset(io, 0, sizeof(*io));
 		atomic_inc(&EXT4_I(inode)->i_ioend_count);
 		io->inode = inode;
-		INIT_WORK(&io->work, ext4_end_io_work);
 		INIT_LIST_HEAD(&io->list);
+		atomic_set(&io->count, 1);
 	}
 	return io;
 }
 
-/*
- * Print an buffer I/O error compatible with the fs/buffer.c.  This
- * provides compatibility with dmesg scrapers that look for a specific
- * buffer I/O error message.  We really need a unified error reporting
- * structure to userspace ala Digital Unix's uerf system, but it's
- * probably not going to happen in my lifetime, due to LKML politics...
- */
-static void buffer_io_error(struct buffer_head *bh)
+void ext4_put_io_end_defer(ext4_io_end_t *io_end)
 {
-	char b[BDEVNAME_SIZE];
-	printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n",
-			bdevname(bh->b_bdev, b),
-			(unsigned long long)bh->b_blocknr);
+	if (atomic_dec_and_test(&io_end->count)) {
+		if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
+			ext4_release_io_end(io_end);
+			return;
+		}
+		ext4_add_complete_io(io_end);
+	}
+}
+
+int ext4_put_io_end(ext4_io_end_t *io_end)
+{
+	int err = 0;
+
+	if (atomic_dec_and_test(&io_end->count)) {
+		if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
+			err = ext4_convert_unwritten_extents(io_end->handle,
+						io_end->inode, io_end->offset,
+						io_end->size);
+			io_end->handle = NULL;
+			ext4_clear_io_unwritten_flag(io_end);
+		}
+		ext4_release_io_end(io_end);
+	}
+	return err;
+}
+
+ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
+{
+	atomic_inc(&io_end->count);
+	return io_end;
 }
 
+/* BIO completion function for page writeback */
 static void ext4_end_bio(struct bio *bio, int error)
 {
 	ext4_io_end_t *io_end = bio->bi_private;
-	struct workqueue_struct *wq;
-	struct inode *inode;
-	unsigned long flags;
-	int i;
+	sector_t bi_sector = bio->bi_iter.bi_sector;
 
 	BUG_ON(!io_end);
-	bio->bi_private = NULL;
 	bio->bi_end_io = NULL;
 	if (test_bit(BIO_UPTODATE, &bio->bi_flags))
 		error = 0;
-	bio_put(bio);
-
-	for (i = 0; i < io_end->num_io_pages; i++) {
-		struct page *page = io_end->pages[i]->p_page;
-		struct buffer_head *bh, *head;
-		int partial_write = 0;
-
-		head = page_buffers(page);
-		if (error)
-			SetPageError(page);
-		BUG_ON(!head);
-		if (head->b_size == PAGE_CACHE_SIZE)
-			clear_buffer_dirty(head);
-		else {
-			loff_t offset;
-			loff_t io_end_offset = io_end->offset + io_end->size;
-
-			offset = (sector_t) page->index << PAGE_CACHE_SHIFT;
-			bh = head;
-			do {
-				if ((offset >= io_end->offset) &&
-				    (offset+bh->b_size <= io_end_offset)) {
-					if (error)
-						buffer_io_error(bh);
-
-					clear_buffer_dirty(bh);
-				}
-				if (buffer_delay(bh))
-					partial_write = 1;
-				else if (!buffer_mapped(bh))
-					clear_buffer_dirty(bh);
-				else if (buffer_dirty(bh))
-					partial_write = 1;
-				offset += bh->b_size;
-				bh = bh->b_this_page;
-			} while (bh != head);
-		}
-
-		put_io_page(io_end->pages[i]);
-
-		/*
-		 * If this is a partial write which happened to make
-		 * all buffers uptodate then we can optimize away a
-		 * bogus readpage() for the next read(). Here we
-		 * 'discover' whether the page went uptodate as a
-		 * result of this (potentially partial) write.
-		 */
-		if (!partial_write)
-			SetPageUptodate(page);
-	}
-	io_end->num_io_pages = 0;
-	inode = io_end->inode;
 
 	if (error) {
-		io_end->flag |= EXT4_IO_END_ERROR;
-		ext4_warning(inode->i_sb, "I/O error writing to inode %lu "
+		struct inode *inode = io_end->inode;
+
+		ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
 			     "(offset %llu size %ld starting block %llu)",
-			     inode->i_ino,
+			     error, inode->i_ino,
 			     (unsigned long long) io_end->offset,
 			     (long) io_end->size,
 			     (unsigned long long)
-			     bio->bi_sector >> (inode->i_blkbits - 9));
+			     bi_sector >> (inode->i_blkbits - 9));
+		mapping_set_error(inode->i_mapping, error);
 	}
 
-	/* Add the io_end to per-inode completed io list*/
-	spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
-	list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
-	spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
-
-	wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
-	/* queue the work to convert unwritten extents to written */
-	queue_work(wq, &io_end->work);
+	if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
+		/*
+		 * Link bio into list hanging from io_end. We have to do it
+		 * atomically as bio completions can be racing against each
+		 * other.
+		 */
+		bio->bi_private = xchg(&io_end->bio, bio);
+		ext4_put_io_end_defer(io_end);
+	} else {
+		/*
+		 * Drop io_end reference early. Inode can get freed once
+		 * we finish the bio.
+		 */
+		ext4_put_io_end_defer(io_end);
+		ext4_finish_bio(bio);
+		bio_put(bio);
+	}
 }
 
 void ext4_io_submit(struct ext4_io_submit *io)
@@ -283,149 +347,151 @@ void ext4_io_submit(struct ext4_io_submit *io)
 		BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
 		bio_put(io->io_bio);
 	}
-	io->io_bio = 0;
-	io->io_op = 0;
-	io->io_end = 0;
+	io->io_bio = NULL;
 }
 
-static int io_submit_init(struct ext4_io_submit *io,
-			  struct inode *inode,
-			  struct writeback_control *wbc,
-			  struct buffer_head *bh)
+void ext4_io_submit_init(struct ext4_io_submit *io,
+			 struct writeback_control *wbc)
+{
+	io->io_op = (wbc->sync_mode == WB_SYNC_ALL ?  WRITE_SYNC : WRITE);
+	io->io_bio = NULL;
+	io->io_end = NULL;
+}
+
+static int io_submit_init_bio(struct ext4_io_submit *io,
+			      struct buffer_head *bh)
 {
-	ext4_io_end_t *io_end;
-	struct page *page = bh->b_page;
 	int nvecs = bio_get_nr_vecs(bh->b_bdev);
 	struct bio *bio;
 
-	io_end = ext4_init_io_end(inode, GFP_NOFS);
-	if (!io_end)
+	bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES));
+	if (!bio)
 		return -ENOMEM;
-	do {
-		bio = bio_alloc(GFP_NOIO, nvecs);
-		nvecs >>= 1;
-	} while (bio == NULL);
-
-	bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
+	bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
 	bio->bi_bdev = bh->b_bdev;
-	bio->bi_private = io->io_end = io_end;
 	bio->bi_end_io = ext4_end_bio;
-
-	io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh);
-
+	bio->bi_private = ext4_get_io_end(io->io_end);
 	io->io_bio = bio;
-	io->io_op = (wbc->sync_mode == WB_SYNC_ALL ?
-			WRITE_SYNC_PLUG : WRITE);
 	io->io_next_block = bh->b_blocknr;
 	return 0;
 }
 
 static int io_submit_add_bh(struct ext4_io_submit *io,
-			    struct ext4_io_page *io_page,
 			    struct inode *inode,
-			    struct writeback_control *wbc,
 			    struct buffer_head *bh)
 {
-	ext4_io_end_t *io_end;
 	int ret;
 
-	if (buffer_new(bh)) {
-		clear_buffer_new(bh);
-		unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
-	}
-
-	if (!buffer_mapped(bh) || buffer_delay(bh)) {
-		if (!buffer_mapped(bh))
-			clear_buffer_dirty(bh);
-		if (io->io_bio)
-			ext4_io_submit(io);
-		return 0;
-	}
-
 	if (io->io_bio && bh->b_blocknr != io->io_next_block) {
 submit_and_retry:
 		ext4_io_submit(io);
 	}
 	if (io->io_bio == NULL) {
-		ret = io_submit_init(io, inode, wbc, bh);
+		ret = io_submit_init_bio(io, bh);
 		if (ret)
 			return ret;
 	}
-	io_end = io->io_end;
-	if ((io_end->num_io_pages >= MAX_IO_PAGES) &&
-	    (io_end->pages[io_end->num_io_pages-1] != io_page))
-		goto submit_and_retry;
-	if (buffer_uninit(bh))
-		io->io_end->flag |= EXT4_IO_END_UNWRITTEN;
-	io->io_end->size += bh->b_size;
-	io->io_next_block++;
 	ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
 	if (ret != bh->b_size)
 		goto submit_and_retry;
-	if ((io_end->num_io_pages == 0) ||
-	    (io_end->pages[io_end->num_io_pages-1] != io_page)) {
-		io_end->pages[io_end->num_io_pages++] = io_page;
-		atomic_inc(&io_page->p_count);
-	}
+	io->io_next_block++;
 	return 0;
 }
 
 int ext4_bio_write_page(struct ext4_io_submit *io,
 			struct page *page,
 			int len,
-			struct writeback_control *wbc)
+			struct writeback_control *wbc,
+			bool keep_towrite)
 {
 	struct inode *inode = page->mapping->host;
-	unsigned block_start, block_end, blocksize;
-	struct ext4_io_page *io_page;
+	unsigned block_start, blocksize;
 	struct buffer_head *bh, *head;
 	int ret = 0;
+	int nr_submitted = 0;
 
 	blocksize = 1 << inode->i_blkbits;
 
+	BUG_ON(!PageLocked(page));
 	BUG_ON(PageWriteback(page));
-	set_page_writeback(page);
+
+	if (keep_towrite)
+		set_page_writeback_keepwrite(page);
+	else
+		set_page_writeback(page);
 	ClearPageError(page);
 
-	io_page = kmem_cache_alloc(io_page_cachep, GFP_NOFS);
-	if (!io_page) {
-		set_page_dirty(page);
-		unlock_page(page);
-		return -ENOMEM;
-	}
-	io_page->p_page = page;
-	atomic_set(&io_page->p_count, 1);
-	get_page(page);
-
-	for (bh = head = page_buffers(page), block_start = 0;
-	     bh != head || !block_start;
-	     block_start = block_end, bh = bh->b_this_page) {
-		block_end = block_start + blocksize;
+	/*
+	 * Comments copied from block_write_full_page:
+	 *
+	 * The page straddles i_size.  It must be zeroed out on each and every
+	 * writepage invocation because it may be mmapped.  "A file is mapped
+	 * in multiples of the page size.  For a file that is not a multiple of
+	 * the page size, the remaining memory is zeroed when mapped, and
+	 * writes to that region are not written out to the file."
+	 */
+	if (len < PAGE_CACHE_SIZE)
+		zero_user_segment(page, len, PAGE_CACHE_SIZE);
+	/*
+	 * In the first loop we prepare and mark buffers to submit. We have to
+	 * mark all buffers in the page before submitting so that
+	 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
+	 * on the first buffer finishes and we are still working on submitting
+	 * the second buffer.
+	 */
+	bh = head = page_buffers(page);
+	do {
+		block_start = bh_offset(bh);
 		if (block_start >= len) {
 			clear_buffer_dirty(bh);
 			set_buffer_uptodate(bh);
 			continue;
 		}
-		ret = io_submit_add_bh(io, io_page, inode, wbc, bh);
+		if (!buffer_dirty(bh) || buffer_delay(bh) ||
+		    !buffer_mapped(bh) || buffer_unwritten(bh)) {
+			/* A hole? We can safely clear the dirty bit */
+			if (!buffer_mapped(bh))
+				clear_buffer_dirty(bh);
+			if (io->io_bio)
+				ext4_io_submit(io);
+			continue;
+		}
+		if (buffer_new(bh)) {
+			clear_buffer_new(bh);
+			unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
+		}
+		set_buffer_async_write(bh);
+	} while ((bh = bh->b_this_page) != head);
+
+	/* Now submit buffers to write */
+	bh = head = page_buffers(page);
+	do {
+		if (!buffer_async_write(bh))
+			continue;
+		ret = io_submit_add_bh(io, inode, bh);
 		if (ret) {
 			/*
 			 * We only get here on ENOMEM.  Not much else
 			 * we can do but mark the page as dirty, and
 			 * better luck next time.
 			 */
-			set_page_dirty(page);
+			redirty_page_for_writepage(wbc, page);
 			break;
 		}
+		nr_submitted++;
+		clear_buffer_dirty(bh);
+	} while ((bh = bh->b_this_page) != head);
+
+	/* Error stopped previous loop? Clean up buffers... */
+	if (ret) {
+		do {
+			clear_buffer_async_write(bh);
+			bh = bh->b_this_page;
+		} while (bh != head);
 	}
 	unlock_page(page);
-	/*
-	 * If the page was truncated before we could do the writeback,
-	 * or we had a memory allocation error while trying to write
-	 * the first buffer head, we won't have submitted any pages for
-	 * I/O.  In that case we need to make sure we've cleared the
-	 * PageWriteback bit from the page to prevent the system from
-	 * wedging later on.
-	 */
-	put_io_page(io_page);
+	/* Nothing submitted - we have to end page writeback */
+	if (!nr_submitted)
+		end_page_writeback(page);
 	return ret;
 }