lupin merge progress

[Colin Watson]

On Thu, Aug 23, 2007 at 01:11:48AM +0100, ago wrote:
> I have uploaded some code to https://code.launchpad.net/~ago/lupin/gutsy
> 
> That includes some casper addons to allow for 2 optional boot
> paramters: find_iso and find_preseed. The first will scan all block
> devices and will try to use the first match to be mounted as /cdrom,
> the second does look for a preseed_file on block devices and it passes
> it on as preseed/file. It also resolves the issue of having to map
> windows drives to linux devices: it uses the same find_path algorithm
> to heuristically find a linux device where some particular file
> resides, this is used to find the device containing the user settings
> to be imported by m-a, and the device on which to create the
> loopinstallation files.

Instead of find_preseed, perhaps we could just move preseed/file
processing earlier? You could then do preseed/file=/cdrom/preseed.cfg or
whatever and wouldn't need the extra option.

> The second piece is a simple init.d script (based on gdm) to
> automatically launch Ubiquity in a barebone X session when the cd is
> started with the boot option '--automatic'. Ubiquity could of course
> also be started within a full gnome desktop. The issue is that if we
> are in automatic mode that's likely because users will first run the
> windows wizard, which will install "stuff" and reboot. If they reboot
> into a desktop environment, almost everyone will be lead to believe
> that the installation is completed when it actually did not even start
> and that might be a bit confusing.

This is cool. I'd like to integrate this into ubiquity proper and add a
boot option for it; it's something we've been asked for a number of
times. Thanks! I agree it's necessary in this case.

> 1) fstab + remounting issues:
> 
> /etc/fstab should probably look something like the following:
> 
> /dev/sdX /host defaults ntfs 0 1
> /host/ubuntu/disks/root.disk / ext3 loop 0 2
> /host/ubuntu/disks/boot /boot auto bind 0 3
> /host/ubuntu/disks/swap.disk none swap sw 0 0
> 
> In the above:
> 
> A) /host has to be handled by /etc/init.d/checkroot.sh before /
> B) / has to be remounted rw, which is an issue when the loop device is
> originally created on a r/o filesystem.

Indeed, ntfs-3g doesn't support 'mount -o remount' anyway ...

> C) swap has to be presented as a swap file and not as a loop device,
> which might interfere with the correct functioning of swap.

OK, I'll see if I can figure out how to do this in partman.

> 3) sysctl settings to improve fs robustness in case of hard reboot. I
> still think that sysctl.conf should simply be set during installation
> and not be handled in a separate package which creates an initr.d
> script to call sysctl. The suggested parameters are:
> 
> vm.dirty_background_ratio=0
> vm.dirty_ratio=0
> vm.dirty_expire_centisecs=1
> vm.dirty_writeback_centisecs=1

Doing this in sysctl.conf remains contrary to policy and we can't do it.
We need an init script for this.

> 5) Make sure that windows line endings (CR+LF) are also supported in
> preseed. People may edit the preseed file from within windows before
> rebooting.

Preseeding was fixed some time back to support this.

> 6) Add needed hd/fs modules to live CD initrd

Aren't they in there already? Oh, maybe not the fuse module, hmm. Should
be easy anyway.

> 7) Image files should be contiguous to avoid performance issues. In
> wubi we used a windows api command to preallocate the space without
> having to zero the full file, one way to replicate that within linux
> might be by using the seek option in dd and then writing the last
> block.

Is that guaranteed to allocate continuous space? Normally, using dd like
that in Linux would create a sparse file which isn't quite the same
thing.

> 8) What to do when the loopinstallation folder is not vergin? My vote
> is for ending the installation immediately, if a user wants to
> overwrite an existing installation s/he has to uninstall and reinstall
> explicitly. This might be annoying if the user is repeating the
> installation because the previous one was interrupted, but it's worth
> to be on the safe side.

OK.

> 9) Not sure how the killall5 bug is going
> (https://bugs.launchpad.net/ubuntu/+source/sysvinit/+bug/87763)

I fixed this, as far as I know.

Cheers,

-- 
Colin Watson                                       [cjwatson at ubuntu.com]