Created: 2005-04-23 by MattZimmerman
NFS filesystems can be used to boot systems, either for use in thin client configurations, or in system recovery. The proposed solution is tied in with the replacement of initrd-tools with initramfs-tools. LTSP already uses a userspace NFS Root setup.
With the planned removal of kernel-level support for NFS root configurations, a userland solution is needed. This solution depends on EarlyUserspace.
Scope and Use Cases
- When booting a thin client, some kind of initramfs needs to be there to query the DHCP server to find the appropriate NFS server to mount the root from, and actually run the appropriate user space tools to mount the filesystem.
- For lights-out facilities, a network bootable kernel with an initramfs is needed that has the appropriate tools required to begin recovery of a failed server.
Diskless distcc or Mosix nodes are another example of a case where an initramfs is needed to bring up a node in the cluster.
- Add ability to specify module loading from the kernel command line and additional module details (IRQ, IO Address, Port) to support older network cards.
Tweak dhcplient3 to deal correctly with NFS Root (to handle RENEW requests)
Data Preservation and Migration
Because there are existing initrd-tools and LTSP setups, have to be careful of all the various platforms for the initramfs solution.
User Interface Requirements
- Swap over NFS or NBD.
To have a common Hoary & ltsp kernel + initramfs, we need to know how we were booted (net or cd or disk)
- We will have a kernel command line parameter, which selects which script to execute.