:: Re: [DNG] lilo development has ende…
トップ ページ
このメッセージを削除
このメッセージに返信
著者: Rick Moen
日付:  
To: dng
題目: Re: [DNG] lilo development has ended
Quoting Adam Borowski (kilobyte@???):

> LILO is anything but 'finished'. It's not 'stable', either


Cry me a river.

> -- even on simple filesystems where it works, it dies horribly the
> moment any of blocks the kernel was written on gets moved.


Important rule: If you don't understand how a piece of software works,
you may break it, and indeed probably will do so. Once again, I will
quote the Zen of LILO from
http://linuxmafia.com/faq/Kernel/zen-of-lilo.html (what I wrote in
_Linux Gazette_ back in 2002):

A lot of people never learned the Zen of LILO:

1. /sbin/lilo (the "map installer") is best thought of as a compiler,
and /etc/lilo.conf as its source code (input).

2. Therefore, if you change /etc/lilo.conf or any of the files it points
to, you must run /sbin/lilo before rebooting, to "recompile".

3. You should always have a "safeboot" stanza in /etc/lilo.conf,
pointing to a known-good kernel image that you never fool with, as a
fallback. This ensures that if, e.g., you compile a new kernel but
accidentally omit console support, you can easily recover.


Complaining that you broke your bootloader because you stupidly moved
one of the things it needs to find, when you knew, or should have known,
that doing so would break the bootloader, is like running out of
gasoline because you failed to refill your tank, and _then_ claiming
that said misadventure proves the car is no good. Anyone who does that
(and fails to grasp what he/she did wrong and use the 'safeboot'
fallback or a live CD to fix it) IMO really has no business
administering a Unix system.

> Or, the first disk gets assigned a different position.


Which happens when exactly? Because you're screwing around with
swapping in and out different HBAs? Well, if you're doing that, see
foregoing: You knew, or should knew, that you are likely to chance the
/dev node assignments among your disks, and therefore are going to have
to review _both_ /etc/fstab and your bootloader, if necessary with a
live CD. That's just life in the big city. People who cannot handle
that are welcome to go back to Super Nintendo, and I will lend them a
virtual hanky to cry into.



> And on anything new or semi-new, you have EFI and thus GPT, meaning LILO is
> outright worthless.


My bicycle won't fly to the moon, either. Worthless!

My newest server box to house linuxmafia.com and unixmercenary.net is a
brand new i7-based system from CompuLab, a really sweet machine --
silent, no moving parts, ultra-low power. And no EFI and no EFT. And
good riddance to that.


> Even if you have an old machine and your BIOS allows forcing a specific
> ordering of disks, say goodbye to LILO unless your setup is really simple.
> Ie, no RAID, no btrfs, no zfs, no fsfs, no fancier options of xfs. No
> encryption of any kind. No LVM.


Works for Me.{tm}

But actually you are totally incorrect to claim that RAID is ruled out.
Back before Debian threw away my lilo setup and put GRUB 0.9 in its
place, I had my server system booting a RAID1 md mirror pair perfectly
fine for at least a decade. You just use the map installer on both
disks with an obvious lilo.conf tweak.
http://www.tldp.org/HOWTO/Boot+Root+Raid+LILO-3.html
You didn't know this? It's been standard documentation for, gosh, about
20 years.


You are also incorrect about the rest of your list, provided that the
/boot filesystem omits the things you list.

Frankly, you should have known this, and avoided the error of making
your ludicruously overbroad claim.