Re: definiing deduplication

["Thomas Schmitt" <scdbackup@gmx.net>]

Hi,

Stefan Monnier wrote:
> Presumably the "backuper" is the sysadmin, i.e. the same (group of)
> person who chose the filesystem, so I'd say yes the "backuper" is
> to blame.

I rather mean the whole complex of system maintainer, users, and
backup software.
But even if there is a qualified entity to be blamed, this does not help
with potential backup problems, because the backup cannot wait until
the culprit feels responsible and finds a way to fix the shortcommings.


> BTW, I can't think of any filesystem that can't do snapshots.  E.g. I use
> snapshots with ext4: just ask your block layer to do the snapshot rather
> than your filesystem (in my case I use LVM snapshots).

It took me a while to find out how the block layer can ensure that a
snapshot is consistent on the filesystem level. The answer is Linux VFS
method super_operations.freeze_fs().
  https://www.kernel.org/doc/html/latest/filesystems/vfs.html
Without it a snapshot on block level would be similar to a filesystem
which was not properly unmounted before power-off.

So there might still filesystems in the Linux kernel which do not support
LVM snapshots in a safe way.

  https://manpages.debian.org/bullseye/util-linux/fsfreeze.8.en.html
states about filesystems with freeze capability:
  "List of these filesystems include (2016-12-18) btrfs, ext2/3/4, f2fs,
   jfs, nilfs2, reiserfs, and xfs."
That's quite comprehensive, but by far not the full list of writable
Linux fileystems.

One would have to dig in the kernel whether others have gained a .freeze_fs
method meanwhile. I looked in the torvalds git repo at the super_operations
structs of exfat, fat, fuse, and udf. None of them offers the freeze_fs
method.


Have a nice day :)

Thomas