[Btrfs-users] Subdirectory snapshots

[Chris Mason]

On Wed, 8 Aug 2007 03:43:57 +0100
Claudio Martins <ctpm at ist.utl.pt> wrote:

> 
>  Hi Chris and everyone else,
> 
>  First let me congratulate you for the nice job on btrfs. I've been
> stress testing it with parallel kernel compiles intermixed with
> snapshot taking, up to about ~100 load avg on a dual core box and
> it's doing quite well :-)

Good to hear ;)

> 
>  I noticed that when creating snapshots it seems to always snapshot
> an entire subvolume, that is if one does:
> 
> test1:/mnt# btrfsctl -s snap1 /mnt/default/dir1/dir2
> 
> Now /mnt/snap1 will be a subvolume containing everything /mnt/default 
> contains, instead of only the "dir2" contents. Is there any plans or
> interest on making snapshotting of subdirectories possible?

That's correct.  I do plan making snapshots per subdirectory, there are
just a few things on the todo list that have to come first.

>  The above feature would come in even more handy if the location of
> snapshots would not be limited to the root of the filesystem, i.e. a
> snapshot could reside alongside normal directories (I'm thinking here
> unprivileged users taking snapshots of parts of their own home dirs
> and such).

Yes

> 
>  I understand that btrfs uses a tree of tree roots to keep track of 
> snapshots/subvolumes, and by reading the design document I'm not sure
> if that would be a showstopper to the above features; despite that,
> since every extent is reference counted it would seem possible to
> implement it.

Basically I'll have to make it possible to have a directory entry
anywhere in the tree point to a tree root.  Before that can happen I
need to unify the inode numbers across the whole FS.  Not a huge task,
but I want to get tree locking and enospc out of the way first.

Thanks,
Chris