gfs2: Prevent direct-I/O write fallback errors from getting lost

[ Upstream commit 43a511c ]

When a direct I/O write falls entirely and falls back to buffered I/O and the
buffered I/O fails, the write failed with return value 0 instead of the error
number reported by the buffered I/O. Fix that.

Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

fs/gfs2/file.c

@@ -935,8 +935,11 @@ static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 		current->backing_dev_info = inode_to_bdi(inode);
 		buffered = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
 		current->backing_dev_info = NULL;
-		if (unlikely(buffered <= 0))
+		if (unlikely(buffered <= 0)) {
+			if (!ret)
+				ret = buffered;
 			goto out_unlock;
+		}
 
 		/*
 		 * We need to ensure that the page cache pages are written to