SquashFS?

[Jeremy Katz]

On Fri, 2005-10-21 at 21:42 +0100, Luciano Miguel Ferreira Rocha wrote:
> On Fri, Oct 21, 2005 at 04:24:13PM -0400, Jeremy Katz wrote:
> > On Fri, 2005-10-21 at 19:33 +0100, Luciano Miguel Ferreira Rocha wrote:
> > > On Fri, Oct 21, 2005 at 08:20:00PM +0200, Thorsten Leemhuis wrote:
> > > > We're working on a kernel-module-standard for fedora-extras. I can
> > > > package unionfs and SqashFS then if that would be enough for the LiveCD.
> > > > But there might be a small problem with SqashFS and the initrd:
> > > 
> > > A kernel-module-standard is a necessity for fedora-extras but not for
> > > unionfs/squashfs, IMO. They're small, and can be compiled on-the-fly,
> > > when creating the live cd iso.
> > 
> > This is a non-starter.  Allowing compilation of stuff adds far too much
> > of a "random" element with regards to what's on the system.
> 
> I don't understand that sentence. A "mklivecd" program already has to:
> 
> 1. create initial root
> 2. install packages
> 3. configure livecd system
> 4. create livecd initrd

Creating the live cd initrd should just happen with the standard tools.
Which means mkinitrd doing it when the kernel package is installed.

> What's the damage of adding:
> 3b. make unionfs squashfs?

See above.

> > > And I don't think the lack of support for a "squashed" initrd is a
> > > problem. I believe initramfs works much better for livecds, anyway. :)
> > 
> > But you need to get the squashfs module into the initrd[1].
> > 
> > Jeremy
> > 
> > [1] Using initrd as a generic term for either initrd or initramfs.  Note
> > that Core has been using initramfs by default since FC3... this is one
> > of the reasons why I say that the changes for live CD at an
> > infrastructure level need to be better integrated.  Peter has made some
> > changes in mkinitrd which will help make this easier and I have the
> > start of the changes for kadischi, but I probably won't get back to
> > finishing them up until after test1
> 
> Well, I'd expected the initrd/initramfs to be cunstom built, anyway.
> Specially for the part of searching for the livecd image (in a cdrom,
> nfs, hdd, whatever).

We're making changes to the mkinitrd infrastructure to allow plugging
things like this in.  This will allow for greater consistency between
various methods of system initrds and avoid having to replicate
everything in fifteen different places.

> And about current mkinitrd, isn't --preload=mod.ko sufficient?

mkinitrd gets run by new-kernel-pkg, so there's not really a chance to
add random arguments

Jeremy