[PATCH] fs/ntfs3: restore the file size when fallocate() fails on a sparse file

[Hengyu Liang]

For a sparse file, ntfs_fallocate() first extends the file with
ntfs_set_size() and then allocates the clusters of the range.

However, when this allocation fails, the function returns without
putting the size back. As of now, a fallocate() that fails with ENOSPC
still leaves the file at the requested size, with a part of the range
allocated.

The issue can be reproduced on a volume mounted with the sparse option:

    truncate -s 16M img
    mkntfs -F -q img
    mount -t ntfs3 -o loop,sparse img /mnt
    dd if=/dev/zero of=/mnt/hole bs=512k count=1
    dd if=/dev/zero of=/mnt/fill bs=4k
    sync
    rm /mnt/hole
    sync
    fallocate -l 1M /mnt/f
    ls -l /mnt/f
    df -k /mnt

The fallocate command fails with "No space left on device". Before this
patch, /mnt/f is then 1048576 bytes long and the available space has
dropped from 912K to 400K. After this patch, /mnt/f is empty and the
available space is still 912K.

This patch will set the size back when the allocation fails, with the
call that FALLOC_FL_KEEP_SIZE already uses.

Fixes: 4342306f0f0d ("fs/ntfs3: Add file operations and implementation")
Signed-off-by: Hengyu Liang <hengyul@xxxxxxxxxx>
---
 fs/ntfs3/file.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/fs/ntfs3/file.c b/fs/ntfs3/file.c
index 95dfe6878860..98fb4116031b 100644
--- a/fs/ntfs3/file.c
+++ b/fs/ntfs3/file.c
@@ -673,8 +673,12 @@ static long ntfs_fallocate(struct file *file, int mode, loff_t vbo, loff_t len)
 				err = attr_data_get_block(ni, vcn, cend - vcn,
 							  &lcn, &clen, &new,
 							  true, NULL, false);
-				if (err)
+				if (err) {
+					/* Do not leave the file extended. */
+					if (new_size > i_size)
+						ntfs_set_size(inode, i_size);
 					goto out;
+				}
 			}
 		}
 
-- 
2.53.0