Alternative Boot Loaders

[Phillip Susi]

Colin Watson wrote:
> The XFS item has been disputed from the start, and I can only repeat
> what I've always said regarding it. The problem here is that XFS skates
> close to the edge of what the POSIX specification allows a file system
> to do; in particular it cannot be forced to write data to disk by any
> action short of unmounting and remounting the file system (or possibly
> remounting the file system read-only, but I seem to remember that that
> was also problematic). GRUB requires that data be physically on disk in
> order for grub-install to work. To answer a frequently asked question
> up-front, 'xfs_freeze -f' at least used not to be sufficient to
> guarantee that data had been written to disk; I tested this extensively
> some time ago.
> 
> This problem is a race condition, and so it is entirely possible that
> some users may never see it, while other users will always see it, and
> still others will randomly see it or not. Installing LILO, whatever its
> user interface deficiencies, is much better than the boot loader
> installation process crashing.

How is lilo not effected by this as well?  It has to get the block list 
of the kernel and write it to the lilo image on disk, so couldn't the 
kernel not be committed to disk yet when lilo runs, thus causing it to 
fail to get its on disk location?