Replying to myself again...
--- Jane Dogalt <jdogalt(a)yahoo.com> wrote:
--- Jesse Keating <jkeating(a)redhat.com> wrote:
> On Thursday 25 January 2007 17:38, Jane Dogalt wrote:
> > I guess I should go read the rest of that thread to figure out
> where
> > the work remains to get a pungi/mock build that doesn't require
any
> > root priveleges...
>
> A way for users to create and mount loopback filesystems for the
> anaconda
> stage1/2 images.
How's this:
I've been deep in the initrd created by pilgrim for livecds...
I've recently read someones post on this list suggesting that an
initrd
could be fleshed out into a fully functional rescue system...
I've been in the initrd pursuing my rebootless installer idea, which
actually looks pretty trivial... (I'm confident it will be done by
fudcon)....
I've been upset at how much time I've spent on various hacks to
provide
input to qemu, when it appears the --append, --kernel, and --initrd
probably will get me what I need in a cleaner way...
AND NOW...
I think I can solve the problem you just stated-
make an initrd as fat or a bit fatter even than the pilgrim livecd,
then use qemu with no system image other than a specified
kernel/initrd
that come from the running system.
Of course, since we no longer need loop/mount/mkfs for initramfs, you
of course just build a custom initramfs based on existing ones, but
adding the tools you need, i.e. mkfs.<whatever> (and libs, as par
pilgrim script)
Or, if you really wanted to be sick, you could just pass the host root
filesystem as a read only device to qemu, then have your custom
initramfs mount it, chroot to it, and work from it mounted read only,
writing data either to /dev/hdb as mentioned, or via qemu's emulated
samba server. But certainly for pungi, this would be overkill, since
all you would need is the short script
qmkfs -t <filesystem type> -o <mkfs opts> -r <src root dir> [-s
transmogrifyscript.sh] <output_filesystem_image>
So it grabs mkfs.<filesystem type> from the host filesystem, copies it
to the new initramfs along with the transmogrifyscript (for changing
ownership/permissions of files and any other arbitrary stuff). Then it
runs qemu with the new initramfs as described, eventually extracting
the output to the file specified.
I suppose if nobody else writes the above shell script first, I'll get
around to it sooner or later...
(I'm actually using a much more convoluted system currently which uses
an entire fedora core installation that was just generated. Until I
was digging deep into the very flexible initramfs in pilgrim (maybe not
much different from core, I haven't looked), I didn't realize I could
practically do everything I needed to do as root in just the initramfs
environment. I want to give credit to Jason
https://www.redhat.com/archives/fedora-devel-list/2007-January/msg01191.html
who combined with my own sick adventures of creating md-raid1 mirrors
underneath dm-snapshots by hand using DavidZ's pilgrim's initramfs
eshell, inspired this design of qmkfs.
-dmc/jdog
____________________________________________________________________________________
Food fight? Enjoy some healthy debate
in the Yahoo! Answers Food & Drink Q&A.
http://answers.yahoo.com/dir/?link=list&sid=396545367