On Fri, May 17, 2002 at 12:29:38PM +0200, uaca@... wrote:
> when ext3 mounts a file system that crashed (was not cleanly umounted) it
> recovers it without the need of the fsck.ext3 helper
> but it seems that jfs needs fsck.jfs to recover it, isn't it?
> so if the filesystem that needs to be recovered has fsck.jfs, then file system
> can't be recovered...
> Am I wrong?
It's correct (actually the program needed is logredo).
One way to work around this is to put the logredo into the initrd.
ext2 has the same problem BTW.
The common assumption is that root stays readable enough that it is at least
possible to read the fsck and anything that is needed to call it (/bin/sh
etc.). But then the kernel is usually in this file system too, so who
guarantees that it isn't corrupted too? lilo actually doesn't need file system
metadata so doesn't depend on a replayed log, but other bootloaders do.