direct-io: Handle O_(D)SYNC AIO

Call generic_write_sync() from the deferred I/O completion handler if
O_DSYNC is set for a write request.  Also make sure various callers
don't call generic_write_sync if the direct I/O code returns
-EIOCBQUEUED.

Based on an earlier patch from Jan Kara <jack@suse.cz> with updates from
Jeff Moyer <jmoyer@redhat.com> and Darrick J. Wong <darrick.wong@oracle.com>.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

5 files changed, 40 insertions, 13 deletions

diff --git a/fs/block_dev.c b/fs/block_dev.c
index c7bda5cd3da..1173a4ee083 100644
--- a/fs/block_dev.c
+++ b/fs/block_dev.c
@@ -1519,7 +1519,7 @@ ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
 
 	blk_start_plug(&plug);
 	ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
-	if (ret > 0 || ret == -EIOCBQUEUED) {
+	if (ret > 0) {
 		ssize_t err;
 
 		err = generic_write_sync(file, pos, ret);
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index 8e686a427ce..4d2eb641714 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -1727,7 +1727,7 @@ static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
 	 */
 	BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
 	BTRFS_I(inode)->last_sub_trans = root->log_transid;
-	if (num_written > 0 || num_written == -EIOCBQUEUED) {
+	if (num_written > 0) {
 		err = generic_write_sync(file, pos, num_written);
 		if (err < 0 && num_written > 0)
 			num_written = err;
diff --git a/fs/cifs/file.c b/fs/cifs/file.c
index 7e36ae34e94..9d0dd952ad7 100644
--- a/fs/cifs/file.c
+++ b/fs/cifs/file.c
@@ -2553,7 +2553,7 @@ cifs_writev(struct kiocb *iocb, const struct iovec *iov,
 		mutex_unlock(&inode->i_mutex);
 	}
 
-	if (rc > 0 || rc == -EIOCBQUEUED) {
+	if (rc > 0) {
 		ssize_t err;
 
 		err = generic_write_sync(file, pos, rc);
diff --git a/fs/direct-io.c b/fs/direct-io.c
index 8b31b9f449f..1782023bd68 100644
--- a/fs/direct-io.c
+++ b/fs/direct-io.c
@@ -266,8 +266,18 @@ static ssize_t dio_complete(struct dio *dio, loff_t offset, ssize_t ret,
 		dio->end_io(dio->iocb, offset, transferred, dio->private);
 
 	inode_dio_done(dio->inode);
-	if (is_async)
+	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;
@@ -1183,11 +1193,6 @@ do_blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
 	}
 
 	/*
-	 * Will be decremented at I/O completion time.
-	 */
-	atomic_inc(&inode->i_dio_count);
-
-	/*
 	 * 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
@@ -1195,11 +1200,33 @@ do_blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
 	 */
 	dio->is_async = !is_sync_kiocb(iocb) && !((rw & WRITE) &&
 		(end > i_size_read(inode)));
-
-	retval = 0;
-
 	dio->inode = inode;
 	dio->rw = rw;
+
+	/*
+	 * For AIO O_(D)SYNC writes we need to defer completions to a workqueue
+	 * so that we can call ->fsync.
+	 */
+	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;
+		}
+	}
+
+	/*
+	 * Will be decremented at I/O completion time.
+	 */
+	atomic_inc(&inode->i_dio_count);
+
+	retval = 0;
 	sdio.blkbits = blkbits;
 	sdio.blkfactor = i_blkbits - blkbits;
 	sdio.block_in_file = offset >> blkbits;
diff --git a/fs/ext4/file.c b/fs/ext4/file.c
index 319c9d26279..3da21945ff1 100644
--- a/fs/ext4/file.c
+++ b/fs/ext4/file.c
@@ -149,7 +149,7 @@ ext4_file_dio_write(struct kiocb *iocb, const struct iovec *iov,
 	ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
 	mutex_unlock(&inode->i_mutex);
 
-	if (ret > 0 || ret == -EIOCBQUEUED) {
+	if (ret > 0) {
 		ssize_t err;
 
 		err = generic_write_sync(file, pos, ret);