fs_lvm without CLVM

[apocas]

Let's try to ditch the CLVM requirement from the fs_lvm drivers.

shared drivers needed to be tweaked, since fs_lvm works with two datastores: a system with type shared and image one with type fs_lvm.

During live migration the shared tm driver is called, shared loads the necessary data and calls the correct tm driver. (basically shared is now a dispatcher'ish for other shared tm drivers)

Theoretically this shouldn't break drivers different than fs_lvm, but it should be tested nevertheless.

bc command was used to do some basic math, a bash guru (which I'm not) should be able to drop this requirement. (frontend should feature the bc command ex: yum install bc)

Only live migrate works for now, offline migration does not work because pre and postmigrate tm scripts are not called during offline migration.
vmm could be tweaked to be lvm aware but then it would break compatibility with other scenarios/setups which is a no go.

[jmelis]

Hi Pedro,

thanks a lot for this PR. We intend to merge it into the master branch and include it into OpenNebula 5.0. We are leaving this open for a while until we can start with this campaign, but I just wanted to let you know that we are definitely interested in including this!

thanks!

[jfontan]

I've merged your changes and made changes in branch feature-4110. Some of the changes done:

• fs_lvm is now a system datastore TM
• Uses lvscan instead of lvs. In Ubuntu the command lvs does create new device files.
• Enabled cold migrations
• dd commands don't use sudo. Using sudo makes the commands fail in nfs shares with root_squash and it is not needed as oneadmin should be in disk group
• Uses qemu-img to create new images. The old method (using dd, skip N megabytes and writting 1 byte) creates images of N megabytes plus one byte. Logical Volumes must be multiple of 512 bytes so the size must be exact