Re: ext3 root partition with kernel-image packages
On Fri, May 03, 2002 at 09:50:09AM -0400, Michael Stone wrote:
> > > yes, tune2fs -j /dev/your_hd
> > Please read my previous message on this topic. Doing that is not enough.
> Unfortunately, that *is* enough, sometimes. (I've got several machines
> sitting around with stock kernels & ext3 that are doing exactly that.)
> What should happen is that the kernel should try to mount the root
> partition with whatever filesystems it knows about, and be happy if it
> mounts ext3, and mount ext2 otherwise. But sometimes this doesn't work,
> and I'm not at all clear why. Playing games with pivot_root is hideous
> and should be (is, sometimes) completely unnecessary.
I think it's the order that that kernel tries the filesystems it knows
about... if it tries ext2 before ext3, and ext2 says - oh yes I know
about this one... ext3 will never get a chance to try!
Now when you compile ext2 and ext3 into the kernel together, they build
system includes ext3 first... but with the -i386 kernels you get ext2 in
the kernel and ext3 as a module... now the initrd sucessfully loads
ext3... but it comes in the filesystems list _after_ ext2. ext3 never
gets a chance to see if it can load the fs.
So what I do is have neither ext2 or ext3 in the kernel and make sure
ext3 appears first in /etc/modules and /etc/mkinitrd/modules.
This requires rolling your own kernel or using an etc/mkinitrd/scripts
script to manually mount the fs and then pivot_root until:
> Mike Stone
To UNSUBSCRIBE, email to email@example.com
with a subject of "unsubscribe". Trouble? Contact firstname.lastname@example.org