Tue, 25 Nov 2003

Attaching the Python debugger to anaconda for a Redhat Fedora (or RH9) install. (Ongoing progress notes and mutterings to oneself.) I am trying to debug a problem with RH Kickstart and its inability to deal with pre-existing software RAID partitions. To do so I am adding a debugger to tag along with Anaconda, which means taking the right stage 2 cramfs image and modifying the image. To mount the image, the command line is mount -t cramfs foo.img /mount/point -o loop=/dev/loop1 . You then have to cp -R the stuff elsewhere to edit it. cp will inform you of various hard links that it is refusing to set. I replaced them manually with symlinks. mkcramfs then rebuilds the damn thing. (Hat tips: Alex for his pdb wrappers and this dude for related info.)

The line from tracing import set_trace; set_trace() needs adding somewhere in there, preferably not before necessary. Later on some notes on why it is that this bug happens. The things I need to find are: how does the install process choose which of the .img files in Redhat/base to mount for stage 2, so I know where to edit, how do set break points for the .pyc procedures I suspect my problems are in, and then how to make a full out new Disc 1 for the new install procedure. That's the big thing for the week. Important question for the moment: you have a KickStart floppy that starts up an NFS install. Which of the images get used to start up anaconda? 1. /floppy/initrd.img? (gzipped ext2 image, usually. zcat through to a file, mount on loopback.) /nfs/RedHat/base/*.img? (cramfs)

More notes from poking around: init in the initrd.img is an ELF executable, and etc/inittab does not exist.