On Tue, Jun 26, 2001 at 08:46:29PM -0700, Chris Tillman wrote:
>
> Well, i tried installing from the 2.2.23 floppies and all went well. That
> was on the exact same partition that 2.3.6 had so much trouble with. BTW, I
> pointed the installer at the exact same folder on my HFS drive as I did for
> the other install. It installed 2.2.19, no errors.
maybe you have a fscked download of the boot-floppies images.
> I also did a fsck on the root partition /dev/sdb4 and it said it was clean.
>
> Then I went back to 2.3.6 and installed on a different partition than I had
> previously. I got the same results, various zcat format, crc, length errors.
> Also an unexpected EOF. It had the same trouble reading the files off the
> HFS drive also.
afaik the powerpc 2.2.19 kernel is the same one that is in potato.
> I tried fsck under 2.3.6, it was not found (no symlink in the new b-f
> apparently), so I tried e2fsck on /dev/sdb5 (which was the partition I was
fsck is a wrapper that tries to invoke the right fsck for whatever
filesystem it detects. we are probably trying to save space by
ditching this wrapper.
> installing 2.3.6 on). It just said
> ext2fs library version out of date! (emphasis not mine, and then)
> e2fsck-1.21-WIP, 14-Jun-2001 ...
> Using EXT2FS Library version 1.19, 13-Jul-2000
bleah yeah e2fslib is fscked, thats why badblocks dies too.
> Then I booted back in under the 2.2.23 b-f, and did fsck -v /dev/sdb5 with
> e2fsck-1.18, (on the partition which I had tried to install from 2.3.6). It
> said
> //dev/sdb5 contains a filesystem with errors, check forced.
> Checking inodes, blocks, and sizes.
> Checking directory structure
> Directory inode 26417, block 0, offset 0: directory corrupted
> Salvage?(y)
[snip nasty filesystem corruption]
> 1 symbolic link (1 fast symbolic link)
> 47 files
>
> I then initialized a 3rd partition /dev/sdb6 with the e2fsck-1.18 and
> rebooted with the 2.3.6 boot-floppy. I mounted sdb6 as root, without
> re-initializing and went straight to Install Operating System. similar
> errors were again generated, bad tar headers, input/output errors.
>
> Interestingly, it's always the modules that have difficulty, the kernel
> installation doesn't give errors.
i think your drivers.tgz tarball is corrupt. suffice it to say i
cannot reproduce any of this (except the broken e2fslib).
> I re-initialized sdb5 using b-f-2.3.6, then went back in with b-f-2.2.23 and
> ran fsck on that partition, it was clean. So it's apparently not mke2fs
> that's having a problem, even though fixing the bad blocks scan would
> probably be a good idea.
it will be as soon as we figure out why the ext2 libraries are getting
screwed.
> I really think there must be some problem with the module installation code,
> some change between b-f-2.2.23 and b-f-2.3.6. Somehow it's fscking up the
> filesystem, writing over . and .. , etc.
the module installation code consists of:
chdir ("/target/tmp")
execlog("zcat /instmnt/drivers.tgz | tar -x")
chdir ("/target/tmp/drivers")
execlog("./install.sh")
(basically).
there is simply no way this code is causing this problem, period.
your drivers.tgz file is CLEARLY corrupt, or busybox is broken, and
from my tests busybox is fine. why your filesystem is getting
corrupted i don't know, its generally not possible for userland code
to cause corruption like that (nothing is touching the raw filesystem
except the kernel and mke2fs).
--
Ethan Benson
http://www.alaska.net/~erbenson/
Attachment:
pgpO4TsbOuira.pgp
Description: PGP signature