initrd in 2.4.19

initrd in 2.4.19

[Brian Kuschak]

I noticed initrd handling has changed recently in the
kernel (init/do_mounts.c).  This new code is causing
me problems.  Has anyone else run into this?

Using the 'new style' (not terminating linuxrc)
/linuxrc runs the following sequence:
  mount -o ro /dev/hda2 /new-root
  cd /new-root
  mkdir initrd
  pivot_root . initrd
  exec chroot . sbin/init dev/console 2>&1
But /linuxrc seems to be runnnig as PID 10, and
sbin/init doesn't start.

Using the 'old style' (linuxrc terminates and kernel
starts init):
  mount -o ro /dev/hda2 /new-root
  cd /new-root
  mkdir initrd
  echo 0x0302 > /proc/sys/kernel/real-root-dev
  exit 0
The mount fails with EBUSY in mount_block_root()
because the root device is already mounted:

Attempting to find a root file system on hda2...
 hda1 hda2
 hda: hda1 hda2
XFS mounting filesystem ide0(3,2)
Starting XFS recovery on filesystem: ide0(3,2) (dev:
3/2)
Ending XFS recovery on filesystem: ide0(3,2) (dev:
3/2)
VFS: busy inodes on changed media.
err=-16, p=ext2, name=/dev/root  <--
VFS: Cannot open root device "" or 03:02
Please append a correct "root=" boot option
Kernel panic: VFS: Unable to mount root fs on 03:02


For the old style to work, I have to not mount the
filesystem at all in my /linuxrc. I'm using
2.4.19-pre7 from bk (1.948)

-Brian


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: initrd in 2.4.19

[Brad Parker]

Brian Kuschak wrote:
>
>I noticed initrd handling has changed recently in the
>kernel (init/do_mounts.c).  This new code is causing
>me problems.  Has anyone else run into this?

heh.  I found that every shred of documentation on this was 1/2 right and
1/2 wrong.

I came to the conclusion that you can either pivot_root or use /linuxrc
but not both.

i.e. if you 'turn on' the special /linuxrc running code you can't pivot
root from it.  it's designed to modload a few modules, enough so the kernel's
own code can mount a file system.  that's it.  and it better return since
the kernel is not done initializing.

I want to use pivot_root, so I don't use the special /linuxrc running
code.  Instead I use "init=/linuxrc" as part of the command line.  If
you do this your first script should work fine (pivot follwed by
chroot).  But to be honest, I don't chroot.  I don't think it's
necessary.

I'd love to hear comments on this.

(actually, looking closer, your 2nd script solves the big problem, which
it the changing of the real root dev #.   I'll bet it would work if it
unmounted /dev/hda2 before returning)

-brad

>
>Using the 'new style' (not terminating linuxrc)
>/linuxrc runs the following sequence:
>  mount -o ro /dev/hda2 /new-root
>  cd /new-root
>  mkdir initrd
>  pivot_root . initrd
>  exec chroot . sbin/init dev/console 2>&1
>But /linuxrc seems to be runnnig as PID 10, and
>sbin/init doesn't start.
>
>Using the 'old style' (linuxrc terminates and kernel
>starts init):
>  mount -o ro /dev/hda2 /new-root
>  cd /new-root
>  mkdir initrd
>  echo 0x0302 > /proc/sys/kernel/real-root-dev
>  exit 0
>The mount fails with EBUSY in mount_block_root()
>because the root device is already mounted:
>
>Attempting to find a root file system on hda2...
> hda1 hda2
> hda: hda1 hda2
>XFS mounting filesystem ide0(3,2)
>Starting XFS recovery on filesystem: ide0(3,2) (dev:
>3/2)
>Ending XFS recovery on filesystem: ide0(3,2) (dev:
>3/2)
>VFS: busy inodes on changed media.
>err=-16, p=ext2, name=/dev/root  <--
>VFS: Cannot open root device "" or 03:02
>Please append a correct "root=" boot option
>Kernel panic: VFS: Unable to mount root fs on 03:02
>
>
>For the old style to work, I have to not mount the
>filesystem at all in my /linuxrc. I'm using
>2.4.19-pre7 from bk (1.948)
>
>-Brian
>
>
>

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

RE: initrd in 2.4.19

[Brian Kuschak]

> I want to use pivot_root, so I don't use the special /linuxrc running
> code.  Instead I use "init=/linuxrc" as part of the command line.  If
> you do this your first script should work fine (pivot follwed by
> chroot).  But to be honest, I don't chroot.  I don't think it's
> necessary.

Using init=/linuxrc allows my first script to work.

However, it's a bit restrictive:  if I want to override the usual boot
sequence and get the kernel to mount the real root and drop into a shell I
can't use this method.  With the 'old-style' script, I can set
"init=/bin/bash".  These get applied AFTER /linuxrc terminates.  If I use
your method, I can't do that.

I realized yesterday that my second script works if I unmount /dev/hda2
before exiting /linuxrc.

Thanks,
Brian

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: initrd in 2.4.19

[David Gibson]

On Thu, May 02, 2002 at 10:13:30AM -0700, Brian Kuschak wrote:
>
> > I want to use pivot_root, so I don't use the special /linuxrc running
> > code.  Instead I use "init=/linuxrc" as part of the command line.  If
> > you do this your first script should work fine (pivot follwed by
> > chroot).  But to be honest, I don't chroot.  I don't think it's
> > necessary.
>
> Using init=/linuxrc allows my first script to work.
>
> However, it's a bit restrictive:  if I want to override the usual boot
> sequence and get the kernel to mount the real root and drop into a shell I
> can't use this method.  With the 'old-style' script, I can set
> "init=/bin/bash".  These get applied AFTER /linuxrc terminates.  If I use
> your method, I can't do that.

Remember that any kernel command line options that the kernel doesn't
recognize get passed to init - as environment variables if they
contain a "=" and as command line arguments otherwise.  So it's fairly
easy to invent a new parameter to control the "real" init and have
your linuxrc interpret it and act accordingly.

--
David Gibson			| For every complex problem there is a
david@gibson.dropbear.id.au	| solution which is simple, neat and
				| wrong.  -- H.L. Mencken
http://www.ozlabs.org/people/dgibson

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/