Description of problem: I have setup an nfs export with the fedora core 6 installation tree. When I try to install a xen host with virt-install -n nfstest -r 256 -f /var/xen/nfstest -s 4 -l nfs:localhost:/export/fedora, virt-install fetches the kernel and ramdisk from the nfs mount and starts anaconda. Anaconda fails however, with "That directory could not be mounted from server". It took me a while to realize that, since anaconda now runs on the xen-host, it can't mount an nfs export from localhost, you have to specify the hostname or the ip of the nfs server. So virt-install -n nfstest -r 256 -f /var/xen/nfstest -s 4 -l nfs:192.168.1.15:/export/fedora works. Installing over FTP or HTTP fails for the same reason. This is not a major issue, but it would be nice if this was specified somewhere in the manpage or on the wiki, or maybe give the user an error if localhost or 127.0.0.1 is specified. Version-Release number of selected component (if applicable): [root@odin ~]# uname -r 2.6.18-1.2849.fc6xen [root@odin ~]# rpm -qa | grep virt libvirt-python-0.1.9-1.fc6 virt-manager-0.2.6-2.fc6 python-virtinst-0.95.0-1.fc6 libvirt-0.1.9-1.fc6 How reproducible: virt-install -n nfstest -r 256 -f /var/xen/nfstest -s 4 -l nfs:localhost:/export/fedora
change QA contact
The NFS url 'nfs:localhost:/export/fedora' gets passed through to Anaconda in the guest OS. Obviously when resolved inside the guest, 'localhost' is now refering to the guest OS, and not the host OS. You need to use the real public hostname, or IP address, and not localhost.