Automatic FS Snapshots

Goal and context

Modern filesystem like BTRFS and ZFS supports a feature known as snapshot (some of them like NILFS2 uses that capability in a even more aggressive manner). Basically, a snapshot is nothing more than coherent "photograph" of a filesystem structures and data blocks pointed by them thus enabling the system administrator to see a filesystem in its *exact* state at the time the snapshot was taken even some files have been deleted/changed/created since the snapshot creation.

Filesystems snapshotting is not a new feature as it is known for quite a long time in the Linux world especially by LVM users. However, one of the areas where filesystems like BTRFS shines is it uses a technique known as Copy-on-Write: while a LVM snapshot physically replicates the stripes (LVM is disconnected from the filesystems lying in the LVM logical volumes and can make no assumptions), CoW filesystems like BTRFS are clever enough to avoid unnecessary data blocks duplication. Thus, at the exact time a snapshot is taken there is no difference between a filesystem and its snapshot meaning the snapshot size is very small. As the time goes on and data blocks are modified, the filesystem will detect alterations attempt and will duplicate all the necessary data blocks to preserve the original ones. On the other side, old and obsolete snapshots must be deleted from time to time to free up the data blocks that have been "locked in the past".

Present proposal pretends to be the "big picture" of integrating snapshotting capabilities of modern CoW filesystems like BTRFS/ZFS in portage. Such an integration will enable sysamdins to easily recover from a disaster coming from a clunky update (or worse, a silent and catatrophic libpng-like update). For the project credibility, it mitigiates the degree of an major issue coming out of the blue. For Funtoo developers, the gain is they do not have to replicate their live filesystem in a chroot or play with virtual machines or jailed tests environments of any kind. In the case of something go wrong, it is simple to do a filesystem rollback just as if nothing has ever happened.

Known Linux distros with the concept

So far Fedora is working on integrating BTRFS with Yum.