H. Peter Anvin
hpa at zytor.com
Fri Jul 30 11:13:00 PDT 2004
Alexander E. Patrakov wrote:
> The "kinit" program distributed with klibc 0.153 still attempts to do a
> pivot_root. That doesn't work. The correct way of going into the main
> system is to mount --bind the root and chroot there, as done in the
> "run-init" program. Could you please copy that logic into "kinit"?
> And please fix the comment at the top - there is "/init", not
> "/sbin/init", in initramfs.
Worse, I think the name /init has been burned by people using it... that
whole thing wasn't phased very well.
My current thinking is to call it /kinit in the patch, and AT THE SAME
TIME add a patch to the kernel to call /kinit if it exists. This means
kinit should finish by calling /init if it exists (this is a real issue
if part of the initramfs comes from the kernel image and part
externally), otherwise do a run-init equivalent and spawn the real root.
This is probably my #1 priority for next week.
More information about the klibc