[pve-devel] create snapshots

[Dietmar Maurer]

> if we don't save vm_state, we need to freeze the filesystem. (qemu-ga call
> xfs_freeze, which works for ext3, ext4, GFS2, XFS, and BTRFS)
> 
> 
> 
> Currently, I'm using directsync or writethrough cache mode to be sure that
> the datas are always on disk.
> (I have a fast nvram journal in my nexenta box).
> I never add problem taking live disk snapshots and rollback them.

Ok, so far so good ;-)

Would it help if we split the "savevm" monitor command into
several smaller command, which we can call individually:

1.) snapshot_start

  * this is the first part of do_savevm() till vm_stop

2.) dump_vm_state (or maybe we do this is step 1)

  * write state to some fd

3.) snapshot_drive

  * simply call bdrv_snapshot_create(bs1, sn) for a single drive

4.) snapshot_end

  * resume the VM (maybe this is a normal resume, so we can use the existing 'cont'
   command)

What do you think?