On Sat, Nov 03, 2007 at 02:38:36AM +0000, Cameron L. Spitzer wrote: > In article <9ldY5-Hs-17@gated-at.bofh.it>, Andrew Sackville-West wrote: > > > I've had to learn my > > way around that having just reconfigured my laptop. The critical item > > is the contents of $ROOT. The value of $ROOT gets set by the kernel > > command line and if it doesn't match, then you have trouble. If you > > change that to the appropriate value you can then 'exit' busybox and > > the boot will carry on. > > I didn't know that. If I hadn't had a working option in GRUB > I would have tried editing the kernel command line next. > I've also rescued Debian by booting Knoppix, mounting stuff, > and running a chroot shell. it turns out you can do a reasonable amount of stuff in that busybox shell and if the system is close to booting, you can get it to go. What I was missing, and desperately wanted, was some kind of text editor. I ended up have to do some funky 'cat'ing of files and text input to tweak an encryption-key script. Not fun without an editor, but you sure can do it... > > > > Once you're up and running, then rebuil the > > initrd's. > > I actually tried that before going to volume labels. Rebuilding > the initrd puts the same old /etc/fstab in the new initrd image. sorry, updating /etc/fstab apparently doesn't go without saying... ;). > That doesn't get you past the udev hang. I guess a more sophisticated > update-initrd would alert you to the difference between the > current mtab and the /etc/fstab contents. There are a handful of issues I've seen with currnet initramsfs-tools that probably need addressing. The one I ran across: the lvm and encrypted root scripts expect /dev/mapper/ names with hyphens in them. I have /dev/mapper/vgcrypt-root (an LVM volume on top of encrypted partition). I prefer to access these things through /dev/vgcrypt/root (or whatever it was), but the scripts are expecting that hyphenated name... it appears to only be documented in the code itself. oh well. it works now... > I've been using Debian for a long time. It's just *weird* to > see anything broken like that. somewhere Linus talks about breaking UUID names as well, on purpose, because he thinks there should be some other persistent naming method and since UUID is generated in user space, its not reliable. I don't know, but watch out for that. A
Attachment:
signature.asc
Description: Digital signature