diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 0c8d97b..3aa0b72 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -848,6 +848,7 @@
 	atomic_t i_ioend_count;	/* Number of outstanding io_end structs */
 	/* current io_end structure for async DIO write*/
 	ext4_io_end_t *cur_aio_dio;
+	atomic_t i_aiodio_unwritten; /* Nr. of inflight conversions pending */
 
 	spinlock_t i_block_reservation_lock;
 
@@ -2119,6 +2120,15 @@
 
 #define in_range(b, first, len)	((b) >= (first) && (b) <= (first) + (len) - 1)
 
+/* For ioend & aio unwritten conversion wait queues */
+#define EXT4_WQ_HASH_SZ		37
+#define ext4_ioend_wq(v)   (&ext4__ioend_wq[((unsigned long)(v)) %\
+					    EXT4_WQ_HASH_SZ])
+#define ext4_aio_mutex(v)  (&ext4__aio_mutex[((unsigned long)(v)) %\
+					     EXT4_WQ_HASH_SZ])
+extern wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
+extern struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
+
 #endif	/* __KERNEL__ */
 
 #endif	/* _EXT4_H */
diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
index 63a7581..ccce8a7 100644
--- a/fs/ext4/extents.c
+++ b/fs/ext4/extents.c
@@ -3174,9 +3174,10 @@
 		 * that this IO needs to convertion to written when IO is
 		 * completed
 		 */
-		if (io)
+		if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
 			io->flag = EXT4_IO_END_UNWRITTEN;
-		else
+			atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
+		} else
 			ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
 		if (ext4_should_dioread_nolock(inode))
 			map->m_flags |= EXT4_MAP_UNINIT;
@@ -3463,9 +3464,10 @@
 		 * that we need to perform convertion when IO is done.
 		 */
 		if ((flags & EXT4_GET_BLOCKS_PRE_IO)) {
-			if (io)
+			if (io && !(io->flag & EXT4_IO_END_UNWRITTEN)) {
 				io->flag = EXT4_IO_END_UNWRITTEN;
-			else
+				atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
+			} else
 				ext4_set_inode_state(inode,
 						     EXT4_STATE_DIO_UNWRITTEN);
 		}
diff --git a/fs/ext4/file.c b/fs/ext4/file.c
index 2e8322c..7b80d54 100644
--- a/fs/ext4/file.c
+++ b/fs/ext4/file.c
@@ -55,11 +55,47 @@
 	return 0;
 }
 
+static void ext4_aiodio_wait(struct inode *inode)
+{
+	wait_queue_head_t *wq = ext4_ioend_wq(inode);
+
+	wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_aiodio_unwritten) == 0));
+}
+
+/*
+ * This tests whether the IO in question is block-aligned or not.
+ * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
+ * are converted to written only after the IO is complete.  Until they are
+ * mapped, these blocks appear as holes, so dio_zero_block() will assume that
+ * it needs to zero out portions of the start and/or end block.  If 2 AIO
+ * threads are at work on the same unwritten block, they must be synchronized
+ * or one thread will zero the other's data, causing corruption.
+ */
+static int
+ext4_unaligned_aio(struct inode *inode, const struct iovec *iov,
+		   unsigned long nr_segs, loff_t pos)
+{
+	struct super_block *sb = inode->i_sb;
+	int blockmask = sb->s_blocksize - 1;
+	size_t count = iov_length(iov, nr_segs);
+	loff_t final_size = pos + count;
+
+	if (pos >= inode->i_size)
+		return 0;
+
+	if ((pos & blockmask) || (final_size & blockmask))
+		return 1;
+
+	return 0;
+}
+
 static ssize_t
 ext4_file_write(struct kiocb *iocb, const struct iovec *iov,
 		unsigned long nr_segs, loff_t pos)
 {
 	struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
+	int unaligned_aio = 0;
+	int ret;
 
 	/*
 	 * If we have encountered a bitmap-format file, the size limit
@@ -78,9 +114,31 @@
 			nr_segs = iov_shorten((struct iovec *)iov, nr_segs,
 					      sbi->s_bitmap_maxbytes - pos);
 		}
+	} else if (unlikely((iocb->ki_filp->f_flags & O_DIRECT) &&
+		   !is_sync_kiocb(iocb))) {
+		unaligned_aio = ext4_unaligned_aio(inode, iov, nr_segs, pos);
 	}
 
-	return generic_file_aio_write(iocb, iov, nr_segs, pos);
+	/* Unaligned direct AIO must be serialized; see comment above */
+	if (unaligned_aio) {
+		static unsigned long unaligned_warn_time;
+
+		/* Warn about this once per day */
+		if (printk_timed_ratelimit(&unaligned_warn_time, 60*60*24*HZ))
+			ext4_msg(inode->i_sb, KERN_WARNING,
+				 "Unaligned AIO/DIO on inode %ld by %s; "
+				 "performance will be poor.",
+				 inode->i_ino, current->comm);
+		mutex_lock(ext4_aio_mutex(inode));
+		ext4_aiodio_wait(inode);
+	}
+
+	ret = generic_file_aio_write(iocb, iov, nr_segs, pos);
+
+	if (unaligned_aio)
+		mutex_unlock(ext4_aio_mutex(inode));
+
+	return ret;
 }
 
 static const struct vm_operations_struct ext4_file_vm_ops = {
diff --git a/fs/ext4/page-io.c b/fs/ext4/page-io.c
index 4e9b0a2..955cc30 100644
--- a/fs/ext4/page-io.c
+++ b/fs/ext4/page-io.c
@@ -32,14 +32,8 @@
 
 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];
-
 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;
@@ -48,9 +42,6 @@
 		kmem_cache_destroy(io_page_cachep);
 		return -ENOMEM;
 	}
-	for (i = 0; i < WQ_HASH_SZ; i++)
-		init_waitqueue_head(&ioend_wq[i]);
-
 	return 0;
 }
 
@@ -62,7 +53,7 @@
 
 void ext4_ioend_wait(struct inode *inode)
 {
-	wait_queue_head_t *wq = to_ioend_wq(inode);
+	wait_queue_head_t *wq = ext4_ioend_wq(inode);
 
 	wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_ioend_count) == 0));
 }
@@ -87,7 +78,7 @@
 	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);
+	wq = ext4_ioend_wq(io->inode);
 	if (atomic_dec_and_test(&EXT4_I(io->inode)->i_ioend_count) &&
 	    waitqueue_active(wq))
 		wake_up_all(wq);
@@ -102,6 +93,7 @@
 	struct inode *inode = io->inode;
 	loff_t offset = io->offset;
 	ssize_t size = io->size;
+	wait_queue_head_t *wq;
 	int ret = 0;
 
 	ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
@@ -126,7 +118,16 @@
 	if (io->iocb)
 		aio_complete(io->iocb, io->result, 0);
 	/* clear the DIO AIO unwritten flag */
-	io->flag &= ~EXT4_IO_END_UNWRITTEN;
+	if (io->flag & EXT4_IO_END_UNWRITTEN) {
+		io->flag &= ~EXT4_IO_END_UNWRITTEN;
+		/* Wake up anyone waiting on unwritten extent conversion */
+		wq = ext4_ioend_wq(io->inode);
+		if (atomic_dec_and_test(&EXT4_I(inode)->i_aiodio_unwritten) &&
+		    waitqueue_active(wq)) {
+			wake_up_all(wq);
+		}
+	}
+
 	return ret;
 }
 
diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index 86b0548..f6a318f 100644
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -833,6 +833,7 @@
 	ei->i_sync_tid = 0;
 	ei->i_datasync_tid = 0;
 	atomic_set(&ei->i_ioend_count, 0);
+	atomic_set(&ei->i_aiodio_unwritten, 0);
 
 	return &ei->vfs_inode;
 }
@@ -4800,11 +4801,21 @@
 	kfree(ext4_feat);
 }
 
+/* Shared across all ext4 file systems */
+wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
+struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
+
 static int __init ext4_init_fs(void)
 {
-	int err;
+	int i, err;
 
 	ext4_check_flag_values();
+
+	for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
+		mutex_init(&ext4__aio_mutex[i]);
+		init_waitqueue_head(&ext4__ioend_wq[i]);
+	}
+
 	err = ext4_init_pageio();
 	if (err)
 		return err;