Initramfs status
[Posted February 5, 2003 by corbet]
One bit of unfinished 2.5 business is "initramfs," the boot-time root
filesystem which is tacked onto the kernel binary image. The plan is to
move much of the initialization-time code out of the kernel and into
initramfs; the result should be a smaller kernel and a safer, more flexible boot
process.
The code to support initramfs has been in the kernel for some time. The
big missing piece has been on the user space side. Before anything useful
can be run in user mode as part of the boot process, there must be a whole
environment to build it in. Attaching the C library to the kernel image
is not an option that would appeal to many, so a special-purpose C library
is needed. That library is "klibc," which has been under development by
Greg Kroah-Hartman and others for some time. klibc provides a minimal set
of standard functions, written with an eye toward portability and small
size.
Greg recently posted an update on klibc.
The library seems to be essentially complete, at least until somebody tries
to do something requiring functions which have not been provided. The
sticking point, at the moment, seems to be a bug in the initramfs unpacking
code. Greg is interested in input from anybody who would like to help
debug that problem. Once that's been ironed out, it is mostly just a
matter of figuring out which boot-time operations should be taken out of
the kernel and moved into a user-space implementation. If that is going to
happen in 2.5, it would be nice if it happened soon; making major changes
to the boot process brings with it a real risk of destabilizing the kernel
for a while.
(
Log in to post comments)