virt-clone --connect=qemu:///system -o testldap -n testldap-lvm -f testldap-lvm/ubuntu-kvm/disk0.qcow2

This clones the VM named testldap to testldap-lvm and put its disk file in the subdirectory testldap-lvm/ubuntu-kvm/. After that I still had to convert this qcow2 image to it’s location in an LV, but that’s not what this post is about.

As the machine is cloned, the MAC address of its virtual NIC is also changed. The ‘source’ VM had 52:54:00:f2:cc:40, the new VM was given 00:16:36:46:34:42. As I booted the new VM I noticed it wouldn’t come up as expected. I couldn’t reach it via the fixed IP that I had given the source VM (even though the source VM was shut down, of course). Closer inspection revealed that the interface name for the NIC in the new VM had changed. I vaguely remembered that Debian-derived distro’s do that: because they don’t want NIC name assignments (eth0, eth1, etc.) to change if a new network adapter is added, they tie a name to a MAC address. And, as noted, the MAC address had indeed changed in the cloning process.

The assignments between MAC and eth? name are recorded in the file /etc/udev/rules.d/70-persistent-net.rules. They are set by the script /lib/udev/write_net_rules, so I removed the execute permissions on that file. However, this was not a clean solution, since it resulted in an error on start up. I found that editing /lib/udev/rules.d/75-persistent-net-generator.rules is a far better solution. Adding the lines

# ignore KVM virtual interfaces
ENV{MATCHADDR}=="52:54:00:*", GOTO="persistent_net_generator_end"
# This seems to be the range used by Xen, but also by virt-clone
ENV{MATCHADDR}=="00:16:36:*", GOTO="persistent_net_generator_end"

seems to do the trick (don’t forget to remove the rules already added in /etc/udev/rules.d/70-persistent-net.rules). Make sure to add them after the lines

# read MAC address
ENV{MATCHADDR}="$attr{address}"

so that the variable MATCHADDR has a value. I documented this solution in the Ubuntu bug report that seemed the most appropriate as well.

This solved one problem. Then the next problem reared its ugly head: Both the source VM and the clone refused to finish their boot process, they kept hanging on the NFS mounts defined in /etc/fstab. The only option mountall gave was to enter the root password (after pressing ESC) or type Crtl-D to continue. Doing the latter resulted in nothing but an infinite wait. In an Ubuntu bug report I found that using DHCP for the network interface would solve the problem. And, indeed it did. However, since I want static IP addresses for my servers this was not a solution that I liked. Much to my surprise the NFS mounts worked perfectly after changing the interface (in /etc/network/interfaces) back to static. I don’t know why, but on both VMs I set the configuration for eth0 from static to dhcp, rebooted, changed it back to static and rebooted again to find the problem solved… Strange!

Update 2009-12-18:
As it turns out, the solution to the mount problem doesn’t always work. I tried it again, but now it failed to work after switchting back from DHCP to a static IP. I guess it has something to do with the lease time of the IP, because in the case I described above there was a night between using the DHCP IP and turning static back on. So somewhere, something needs to time out before switching back from DHCP to static IPs works again.