[Ovirt-devel] [PATCH node] update method used to determine boot partition for install

[Joey Boggs]

Alan Pevec wrote:
> On Fri, Jul 31, 2009 at 5:12 PM, Joey Boggs <jboggs at redhat.com 
> <mailto:jboggs at redhat.com>> wrote:
>
>     This will facilitate the switch to readonly rootfs. The current
>     method will not work when symlinking /etc/mtab to /proc/mounts as
>     the udev label /BOOT is used rather than /dev/sda1 or similar. The
>     same output is generated and has been verified to install correctly.
>
>
> We need this symlink even if /etc/mtab is listed in /etc/rwtab ?
There's some issues with the rwtab in this scenario, for mtab to be in 
there the real problem is the lock file created as 
/etc/mtab~$randomnumber can't be written in /etc. there are a few 
bugzillas concerning this but not likely to change anything until Fedora 
12+ timeframe since it requires updating alot of different programs 
mount.* and so forth. If you just add /etc/ then you have multiple 
overlaps and mounts including the items in /config/files and those in 
/etc/rwtab. /etc is then essentially handled as a separate partition 
which is not the most ideal situation and will cause the machine to hang 
up on boot since it can't unmount /etc and requires a manual reboot.  
Been working on a few scenarios for awhile :). I can edit the eval 
statement and make sure that /boot is mounted as well.

>
>         # check that /boot mounted ok and find partition number for GRUB
>     -    eval $(mount|awk '$3 == "/boot" {
>
>     +    eval $(readlink -f /dev/disk/by-label/BOOT|awk {'
>
>
> This wouldn't check that /boot actually mounted successfully.
>
> -        print "disk=" substr($1,1,length($1)-1);
> -        print "disk2=" substr($1,1,length($1)-2);
> -        partN=substr($1,length($1),1); partN--;
> -        print "partN=" partN;
> -    }')
> +    print "disk=" substr($1,1,length($1)-1);
> +    print "disk2=" substr($1,1,length($1)-2);
> +    partN=substr($1,length($1),1); partN--;
> +    print "partN=" partN;
> +    '})
>
> This whitespace only change obscures the real change, please put in a 
> separate patch, but IMHO fixing indentation is not need here.
>