Re: [gentoo-releng] x86-livecd - bugs and suggestions

[Martin Schlemmer <azarah@gentoo.org>]

[-- Attachment #1: Type: text/plain, Size: 3593 bytes --]

On Fri, 2004-01-23 at 20:43, Brad House wrote:
> Commenting inline below.
> 
> >> Bugs:
> >> - Calculation of module dependencies fails at bootup
> >
> > Anybody have a log?
> 
> reason for this is that /lib/modules/${KV} is off a read-only
> filesystem.


Right, as I suggested in my mail to beejay.

>   It should not need to depmod that directory off
> a livecd, as we won't have removed modules, etc.  An alternative
> approach would be to mount /lib/modules/${KV} on a tmpfs,
> make sure the sub directories are symlinks to the read-only
> fs, and copy the files like  modules.pcimap to the tmpfs
> so they are writable...
> 
> The first approach of just skipping the depmod if booting off
> a livecd is the better approach.  Also, this failure really
> doesn't mean anything or do anything harmful, which is one
> reason I didn't fix it in my baselayout.
> 

Easiest will be with a check to CDBOOT.

> >
> >> - selection of keymap not possible
> >
> > We could add an alternative keymap script, that do that, and only
> > install with USE=livecd ... anybody got a template, etc to start
> > off?
> 
> this I don't have a clue about, so i'm not commenting on.
> 

Same - I use the defaults 8)

> >
> >> - Inability to halt/reboot (system just stops when all processes are
> >>   shut-down)
> >
> > Logs if possible?
> 
> The reason for this is quite simple.  The livecd was built with
> the cdfs/loop type as 'normal' which creates a loopback filesystem.
> The loopback filesystem gets unmounted '_before_' everything else
> is remounted readonly.  But since the loopback is readonly as well,
> it should never actually be being umounted before the reboot.  The
> actual / filesystem is a tmpfs with symlinks back to the mounted
> loopback in this scenerio, so when you umount the loopback, the /
> filesystem has symlinks to nothing, hence the hang.
> 
> This is not an issue if you use cdfs/loop type of zisofs or noloop
> as there is no loop device to be umounted.
> 

Ok, I changed things a bit, so might have expected I guess.  I do
however have an issue with the naming.  Some time ago it was decided
that it should be /mnt/livecd.  Now it changed again.  Can we decide
on something now?  If so, fix is easy.

> >
> >> - Hardware-detection not included
> >>
> >
> > Sure will be nifty, but I guess catalyst should just install appropriate
> > package and add to boot runlevel?  Do not see any special handling, or
> > am I missing something?
> 
> I had not gotten a chance to put together the  livecd-tools package.
> That script, as well as net-setup is in there.  Basically, it will just
> be a package that needs to be a dep in the livecd-stage1 packages,
> and a quick one-liner in the default-runscript.sh will get it all
> up and running, really a simple thing.
> 
> >
> >
> > As for 2.6 - think we can start to use module-init-tools as default for
> > ALL profiles, or is there ARCH issues I am not aware of?  Should just
> > be needed to adjust the default virtual.
> 
> Right, as long as the current module-init-tools ebuild installs the
> insmod.old for 2.4 kernels.  I believe at one time it did not.
> 
> 
> Even though I've stepped away from livecd stuff, I need to fix this
> stuff to build my amd64 2004.0 release anyhow.  I'd take care of this
> since I'm more familiar with the problems if anyone wants me to. I can
> do that this weekend.
> 

Sound good.


-- 

Martin Schlemmer
Gentoo Linux Developer, Desktop/System Team Developer
Cape Town, South Africa



[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]