direct-io: move aio_complete into ->end_io

Filesystems with unwritten extent support must not complete an AIO request
until the transaction to convert the extent has been commited.  That means
the aio_complete calls needs to be moved into the ->end_io callback so
that the filesystem can control when to call it exactly.

This makes a bit of a mess out of dio_complete and the ->end_io callback
prototype even more complicated.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Alex Elder <aelder@sgi.com>

fs/direct-io.c

@@ -218,7 +218,7 @@ static struct page *dio_get_page(struct dio *dio)
  * filesystems can use it to hold additional state between get_block calls and
  * dio_complete.
  */
-static int dio_complete(struct dio *dio, loff_t offset, int ret)
+static int dio_complete(struct dio *dio, loff_t offset, int ret, bool is_async)
 {
 	ssize_t transferred = 0;
 
@@ -239,14 +239,6 @@ static int dio_complete(struct dio *dio, loff_t offset, int ret)
 			transferred = dio->i_size - offset;
 	}
 
-	if (dio->end_io && dio->result)
-		dio->end_io(dio->iocb, offset, transferred,
-			    dio->map_bh.b_private);
-
-	if (dio->flags & DIO_LOCKING)
-		/* lockdep: non-owner release */
-		up_read_non_owner(&dio->inode->i_alloc_sem);
-
 	if (ret == 0)
 		ret = dio->page_errors;
 	if (ret == 0)
@@ -254,6 +246,17 @@ static int dio_complete(struct dio *dio, loff_t offset, int ret)
 	if (ret == 0)
 		ret = transferred;
 
+	if (dio->end_io && dio->result) {
+		dio->end_io(dio->iocb, offset, transferred,
+			    dio->map_bh.b_private, ret, is_async);
+	} else if (is_async) {
+		aio_complete(dio->iocb, ret, 0);
+	}
+
+	if (dio->flags & DIO_LOCKING)
+		/* lockdep: non-owner release */
+		up_read_non_owner(&dio->inode->i_alloc_sem);
+
 	return ret;
 }
 
@@ -277,8 +280,7 @@ static void dio_bio_end_aio(struct bio *bio, int error)
 	spin_unlock_irqrestore(&dio->bio_lock, flags);
 
 	if (remaining == 0) {
-		int ret = dio_complete(dio, dio->iocb->ki_pos, 0);
-		aio_complete(dio->iocb, ret, 0);
+		dio_complete(dio, dio->iocb->ki_pos, 0, true);
 		kfree(dio);
 	}
 }
@@ -1126,7 +1128,7 @@ direct_io_worker(int rw, struct kiocb *iocb, struct inode *inode,
 	spin_unlock_irqrestore(&dio->bio_lock, flags);
 
 	if (ret2 == 0) {
-		ret = dio_complete(dio, offset, ret);
+		ret = dio_complete(dio, offset, ret, false);
 		kfree(dio);
 	} else
 		BUG_ON(ret != -EIOCBQUEUED);

fs/ext4/inode.c

@@ -3775,7 +3775,8 @@ static ext4_io_end_t *ext4_init_io_end (struct inode *inode, gfp_t flags)
 }
 
 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
-			    ssize_t size, void *private)
+			    ssize_t size, void *private, int ret,
+			    bool is_async)
 {
         ext4_io_end_t *io_end = iocb->private;
 	struct workqueue_struct *wq;
@@ -3784,7 +3785,7 @@ static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
 
 	/* if not async direct IO or dio with 0 bytes write, just return */
 	if (!io_end || !size)
-		return;
+		goto out;
 
 	ext_debug("ext4_end_io_dio(): io_end 0x%p"
 		  "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
@@ -3795,7 +3796,7 @@ static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
 	if (io_end->flag != EXT4_IO_UNWRITTEN){
 		ext4_free_io_end(io_end);
 		iocb->private = NULL;
-		return;
+		goto out;
 	}
 
 	io_end->offset = offset;
@@ -3812,6 +3813,9 @@ static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
 	list_add_tail(&io_end->list, &ei->i_completed_io_list);
 	spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
 	iocb->private = NULL;
+out:
+	if (is_async)
+		aio_complete(iocb, ret, 0);
 }
 
 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)

fs/ocfs2/aops.c

@@ -578,7 +578,9 @@ bail:
 static void ocfs2_dio_end_io(struct kiocb *iocb,
 			     loff_t offset,
 			     ssize_t bytes,
-			     void *private)
+			     void *private,
+			     int ret,
+			     bool is_async)
 {
 	struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
 	int level;
@@ -592,6 +594,9 @@ static void ocfs2_dio_end_io(struct kiocb *iocb,
 	if (!level)
 		up_read(&inode->i_alloc_sem);
 	ocfs2_rw_unlock(inode, level);
+
+	if (is_async)
+		aio_complete(iocb, ret, 0);
 }
 
 /*

fs/xfs/linux-2.6/xfs_aops.c

@@ -1406,7 +1406,9 @@ xfs_end_io_direct(
 	struct kiocb	*iocb,
 	loff_t		offset,
 	ssize_t		size,
-	void		*private)
+	void		*private,
+	int		ret,
+	bool		is_async)
 {
 	xfs_ioend_t	*ioend = iocb->private;
 
@@ -1452,6 +1454,9 @@ xfs_end_io_direct(
 	 * against double-freeing.
 	 */
 	iocb->private = NULL;
+
+	if (is_async)
+		aio_complete(iocb, ret, 0);
 }
 
 STATIC ssize_t

fs/xfs/linux-2.6/xfs_aops.h

@@ -37,6 +37,8 @@ typedef struct xfs_ioend {
 	size_t			io_size;	/* size of the extent */
 	xfs_off_t		io_offset;	/* offset in the file */
 	struct work_struct	io_work;	/* xfsdatad work queue */
+	struct kiocb		*io_iocb;
+	int			io_result;
 } xfs_ioend_t;
 
 extern const struct address_space_operations xfs_address_space_operations;

include/linux/fs.h

@@ -415,7 +415,8 @@ struct buffer_head;
 typedef int (get_block_t)(struct inode *inode, sector_t iblock,
 			struct buffer_head *bh_result, int create);
 typedef void (dio_iodone_t)(struct kiocb *iocb, loff_t offset,
-			ssize_t bytes, void *private);
+			ssize_t bytes, void *private, int ret,
+			bool is_async);
 
 /*
  * Attribute flags.  These should be or-ed together to figure out what