On Sat, 01 Jun 2013 20:54:15 +0100, Richard W.M. Jones wrote:
On Sat, Jun 01, 2013 at 02:27:50PM +0000, Gabriel de Perthuis wrote:
Hello, As I understand it guestfs appliances normally work as servers and run high-level commands from some external channel.
This is the normal architecture when you're using libguestfs to access a VM or disk image:
http://libguestfs.org/guestfs.3.html#architecture
But it might be possible to bundle a guestfish script to run commands from and do away with the external system. That would make it useful on non-virtualised systems.
It's worth noting you can edit any device, including host devices directly. For example on my laptop:
It's also possible to get libguestfs to connect to an arbitrary socket, so-called 'libguestfs live'. That could include a guestfsd daemon running on the local host. Run the daemon and set LIBGUESTFS_BACKEND=unix:/path/to/some/socket
http://libguestfs.org/guestfs.3.html#backend https://rwmj.wordpress.com/2011/07/07/how-does-libguestfs-live-work/#content
It would also be possible to run qemu-nbd or sshd on a physical machine and access it using libguestfs (>= 1.22) from another machine. Indeed this is a possible architecture under discussion for the new version of virt-v2v.
Hope that gives you some ideas. If you have a specific use scenario you are interested in, then please let us know.
I was thinking of running some operations on the root filesystem by kexec-rebooting into an appliance to get exclusive access. I would like to do something like that for my blocks tool (which does conversions to LVM and bcache).
supermin + supermin-helper + blocks might be useful to get this to work (or another image builder like Oz, but I haven't played with this much). I'm not yet sure I should integrate with guestfs, but it might be useful for VMs. If so, libguestfs-live could be used in the kexec case.