[9fans] the perpetual disk partition problem

[9fans] the perpetual disk partition problem

[ron minnich]

Well plan 9 is now booting under lguest, it's even faster than xen.

Now I'm up to disk io, and am back at the usual '9load sets up
partitions, right?' step, and have no partitions. Now, I had proposed
a few years back putting that stuff into boot, but that proposal was
seen to be bad form.

So, question, what are you folks doing to partition disks nowadays
when 9load has not loaded your kernel? I'm curious.

ron

Re: [9fans] the perpetual disk partition problem

[erik quanstrom]

silly ideas:

1. couldn't you just use a kfs in a fdisk/fat partition?

2. another approach would be to use russ' boot-to-rc script
trick.

3. prep the disk on a plan 9 machine.

i was fighting a similar problem this morning.  in my case
the problem is that it is quite difficult to convince 9load
to do any network io in addition to bootp/tftp.  this makes
finding a partition table on an aoe target somewhat difficult.

here's my solution.  in plan9.ini, add a line like

sdf0part=9fat 0 20480/nvram 20480 20481/fossil 20481 41963521

where this is just the output from prep -p reformatted.
disgusting, no?  and it doesn't even solve your problem, because you
don't have established parititions.

- erik

Re: [9fans] the perpetual disk partition problem

[erik quanstrom]

hey, it wouldn't be too hard to port prep to linux, right?

- erik

Re: [9fans] the perpetual disk partition problem

[Steve Simon]

> So, question, what are you folks doing to partition disks nowadays
> when 9load has not loaded your kernel? I'm curious.

You could just boot the live CDROM to install onto a physical disk
and then boot your kernel under lguest and point it at this, rather than
a virtual disk in a linux file. I suspose you could even use Linux dd to copy
this disk to a linux file and then use that.

-Steve

Re: [9fans] the perpetual disk partition problem

[ron minnich]

On 7/17/07, Steve Simon <steve@quintile.net> wrote:
> > So, question, what are you folks doing to partition disks nowadays
> > when 9load has not loaded your kernel? I'm curious.
>
> You could just boot the live CDROM to install onto a physical disk
> and then boot your kernel under lguest and point it at this, rather than
> a virtual disk in a linux file. I suspose you could even use Linux dd to copy
> this disk to a linux file and then use that.

yeah but I want to do the whole thing in lguest, as a test.

The boot cd does not help me -- cd is partitioned. I think the
question still needs a good answer -- you have a cd and a disk image,
you boot plan 9 without benefit of 9load, how do you get #S/sdD0/data
set up? you need to write to ctl. What's the sensible way to do this?

ron

Re: [9fans] the perpetual disk partition problem

[erik quanstrom]

> The boot cd does not help me -- cd is partitioned. I think the
> question still needs a good answer -- you have a cd and a disk image,
> you boot plan 9 without benefit of 9load, how do you get #S/sdD0/data
> set up? you need to write to ctl. What's the sensible way to do this?
> 
> ron

why can't you just put the (known) offsets in your plan9.ini like i am doing
for aoe?  it's distasteful, but it does get the job done.

if you don't like that you could put fdisk and prep in your kernel
image, write a new boot method, localpart, that partitions before
it calls local.  you could import the code for this from 9load.

- erik

Re: [9fans] the perpetual disk partition problem

[ron minnich]

On 7/17/07, erik quanstrom <quanstro@coraid.com> wrote:

> why can't you just put the (known) offsets in your plan9.ini like i am doing
> for aoe?  it's distasteful, but it does get the job done.
>

Until the offsets change. This is intended to be used by people other than me.

> if you don't like that you could put fdisk and prep in your kernel
> image, write a new boot method, localpart, that partitions before
> it calls local.  you could import the code for this from 9load.

That's what I did several years ago for the xen 2.0 port but there was
an objection to it that I never understood. I'll go with that.

ron

Re: [9fans] the perpetual disk partition problem

[erik quanstrom]

On Tue Jul 17 13:08:26 EDT 2007, rminnich@gmail.com wrote:
> On 7/17/07, erik quanstrom <quanstro@coraid.com> wrote:
> 
> > why can't you just put the (known) offsets in your plan9.ini like i am doing
> > for aoe?  it's distasteful, but it does get the job done.
> >
> 
> Until the offsets change. This is intended to be used by people other than me.

there always has to be a catch.... ;-)

can you generate these offsets as part of the install?

> 
> > if you don't like that you could put fdisk and prep in your kernel
> > image, write a new boot method, localpart, that partitions before
> > it calls local.  you could import the code for this from 9load.
> 
> That's what I did several years ago for the xen 2.0 port but there was
> an objection to it that I never understood. I'll go with that.

it seems like a good thing to keep the setup out of the kernel proper,
but any sort of sd device that needs to use the network is going to be difficult
with the current 9load.  perhaps with a few tweaks it could be
much easier.

- erik

Re: [9fans] the perpetual disk partition problem

[Uriel]

> > if you don't like that you could put fdisk and prep in your kernel
> > image, write a new boot method, localpart, that partitions before
> > it calls local.  you could import the code for this from 9load.
>
> That's what I did several years ago for the xen 2.0 port but there was
> an objection to it that I never understood. I'll go with that.

I can't think of why this might be a bad idea, boot certainly has some
limitations (like only allowing a small number of files), but if that
is a problem we should remove those limitations, it seems the ideal
place to put this kind of things.

If we are ever going to kill 9load (and I'm still intending to
eventually do it), putting stuff in boot seems to be the only
reasonable option, and a rather good one I might add.

uriel

P.S.: As a way to bypass the current limitations one could put a paqfs
into boot.