dmsquash-live-root + NFS?

dmsquash-live-root + NFS?

[Scott Olsen]

First of all, many thanks for all the work on Dracut. We're moving our
network clients over to Dracut from custom mkinitrd build environment,
and wow, what an improvement.

What we are looking to do is: the exact execution path of
dmsquash-live (with an ext3fs.img inside of squashfs) and copy to
live_ram but with the source of the image being a NFS mount instead of
a block device.

We hacked it to work, but I want to know the proper way to put this
into the dracut framework/architecture.

What we have now is:

- inside cmdline/parse-dmsquash-live, we set netroot to be the nfs server

- let everything proceed as normal

- then at the end of /sbin/nfsroot we just call
/sbin/dmsquash-live-root and everything works (after a few tweaks)
since the NFS mount is already mounted on NEWROOT and live_dir points
to the squashfs image.

--

Like I said, its up and working but is there a better way to implement
this?  any architecture pointers would be greatly appreciated!!

Is there any bad effect of having the udev rules created by
dmsquash-live-genrules run along with the netroot/nfsroot code run?

We had to disable the initqueue check_finished test in init because
init would break after the retry limit saying no root device found
(it's there but does not pass the test).  Is this a consequence of
having both dmsquash-live and nfsroot?


thanks

Scott

Re: dmsquash-live-root + NFS?

[Harald Hoyer]

On 06/04/2010 03:35 AM, Scott Olsen wrote:
> First of all, many thanks for all the work on Dracut. We're moving our
> network clients over to Dracut from custom mkinitrd build environment,
> and wow, what an improvement.
>
> What we are looking to do is: the exact execution path of
> dmsquash-live (with an ext3fs.img inside of squashfs) and copy to
> live_ram but with the source of the image being a NFS mount instead of
> a block device.
>
> We hacked it to work, but I want to know the proper way to put this
> into the dracut framework/architecture.
>
> What we have now is:
>
> - inside cmdline/parse-dmsquash-live, we set netroot to be the nfs server
>
> - let everything proceed as normal
>
> - then at the end of /sbin/nfsroot we just call
> /sbin/dmsquash-live-root and everything works (after a few tweaks)
> since the NFS mount is already mounted on NEWROOT and live_dir points
> to the squashfs image.
>
> --
>
> Like I said, its up and working but is there a better way to implement
> this?  any architecture pointers would be greatly appreciated!!
>
> Is there any bad effect of having the udev rules created by
> dmsquash-live-genrules run along with the netroot/nfsroot code run?
>
> We had to disable the initqueue check_finished test in init because
> init would break after the retry limit saying no root device found
> (it's there but does not pass the test).  Is this a consequence of
> having both dmsquash-live and nfsroot?
>
>
> thanks
>
> Scott

Care to show your modifications, so I can get a better feeling what you are doing?