On Sat, 8 Mar 2025, Thomas Schmitt wrote:
But if, instead, what you want is the backing store to grow then use a
sparse file as the backing store.
This idea reduces the search to filesystems which don't hop around and
write without need to new random places on their storage medium.
# truncate -s 1T file
# mke2fs -t ext4 file
That's a 1T ext4 filesystem that is, when empty, using around 20M of
disk space.
Question is how its "du" size grows when you put a few files into it,
overwrite their content, rename and copy them.
# fallocate -l 100M file
# du -h file
101M file
# mke2fs -t ext4 -E nodiscard file
mke2fs 1.47.0 (5-Feb-2023)
Creating filesystem with 102400 1k blocks and 25584 inodes
Filesystem UUID: 7e49b294-4c51-407a-9b84-557755c7baed
Superblock backups stored on blocks:
8193, 24577, 40961, 57345, 73729
Allocating group tables: done
Writing inode tables: done
Creating journal (4096 blocks): done
Writing superblocks and filesystem accounting information: done
# du -h file
101M file
# mount -o loop file mnt
# fstrim -v mnt
mnt: 88.2 MiB (92486656 bytes) trimmed
# umount mnt
# du -h file
5.7M file
It can be cleaned up periodically. Ideally you'd want a fuse fs that
supported fstrim I guess. I don't know whether this is even possible.
ext4 fuse doesn't support SEEK_HOLE/SEEK_DATA (at least not last time I
looked) so I doubt it will support discard.
