Re: pivot_root alternative

Re: pivot_root alternative

[Salvatore De Paolis]

On Mon, 25 Jun 2007 14:19:02 +0400
Michael Tokarev <mjt@tls.msk.ru> wrote:

> Salvatore De Paolis wrote:
> > 
> >> use run-init from klibc?
> >  
> > Thank you both, i'll take a look at run-init:)
> 
> By the way, busybox now includes similar applet,
> named switch_root.
> 
> /mjt

yeah, i saw it yesterday after a while. Anyway the problem got a solution,
everything seems to work just fine.

Regards
S.

pivot_root alternative

[Salvatore De Paolis]

hi all

i built the kernel with a busybox initramfs which runs from a usb stick. While
i boot i need to mount the real root and i'm using pivot_root.
Pivot_root . old-root return to me an "Invalid argument" error and googling i
found here http://lkml.org/lkml/2005/10/8/17 that pivot_root should be avoided.
I want to ask now what i should do to switch to a real root?

Regards
S.

Re: pivot_root alternative

[Jan Engelhardt]

On Jun 24 2007 23:27, Salvatore De Paolis wrote:
>
>i built the kernel with a busybox initramfs which runs from a usb stick. While
>i boot i need to mount the real root and i'm using pivot_root.
>Pivot_root . old-root return to me an "Invalid argument" error and googling i
>found here http://lkml.org/lkml/2005/10/8/17 that pivot_root should be avoided.
>I want to ask now what i should do to switch to a real root?

use run-init from klibc?


	Jan
-- 

Re: pivot_root alternative

[Arnd Hannemann]

Jan Engelhardt schrieb:
> On Jun 24 2007 23:27, Salvatore De Paolis wrote:
>> i built the kernel with a busybox initramfs which runs from a usb stick. While
>> i boot i need to mount the real root and i'm using pivot_root.
>> Pivot_root . old-root return to me an "Invalid argument" error and googling i
>> found here http://lkml.org/lkml/2005/10/8/17 that pivot_root should be avoided.
>> I want to ask now what i should do to switch to a real root?
> 
> use run-init from klibc?

However be aware of the limitation of run-init:
It only works for direct childs of root directory
/mnt/newroot won't work
/newroot should work ;-)

Good luck,
Arnd

Re: pivot_root alternative

[Salvatore De Paolis]

On Sun, 24 Jun 2007 23:39:25 +0200
Arnd Hannemann <arnd@arndnet.de> wrote:

> use run-init from klibc?
 
> However be aware of the limitation of run-init:
> It only works for direct childs of root directory
> /mnt/newroot won't work
> /newroot should work ;-)
> 
> Good luck,
> Arnd

Thank you both, i'll take a look at run-init:)
Nice to know about the child limitation but i just have to find klibc-utils
inside buildroot or busybox if even exists:)

Regards
S.

Re: pivot_root alternative

[Michael Tokarev]

Salvatore De Paolis wrote:
> 
>> use run-init from klibc?
>  
> Thank you both, i'll take a look at run-init:)

By the way, busybox now includes similar applet,
named switch_root.

/mjt

Re: pivot_root alternative

[H. Peter Anvin]

Arnd Hannemann wrote:
> Jan Engelhardt schrieb:
>> On Jun 24 2007 23:27, Salvatore De Paolis wrote:
>>> i built the kernel with a busybox initramfs which runs from a usb stick. While
>>> i boot i need to mount the real root and i'm using pivot_root.
>>> Pivot_root . old-root return to me an "Invalid argument" error and googling i
>>> found here http://lkml.org/lkml/2005/10/8/17 that pivot_root should be avoided.
>>> I want to ask now what i should do to switch to a real root?
>> use run-init from klibc?
> 
> However be aware of the limitation of run-init:
> It only works for direct childs of root directory
> /mnt/newroot won't work
> /newroot should work ;-)
> 

Hm, that should probably be fixed.  It should be able to work on any
directory as long as there isn't another *filesystem* involved...

	-hpa

Re: pivot_root alternative

[Kyle Moffett]

On Jun 24, 2007, at 17:27:34, Salvatore De Paolis wrote:
> i built the kernel with a busybox initramfs which runs from a usb  
> stick. While i boot i need to mount the real root and i'm using  
> pivot_root.  Pivot_root . old-root return to me an "Invalid  
> argument" error and googling i found here http://lkml.org/lkml/ 
> 2005/10/8/17 that pivot_root should be avoided.  I want to ask now  
> what i should do to switch to a real root?

One interesting fact about the mount() syscall is this: when  
overmounting a directory tree, a process whose current working  
directory is within said directory tree will see the old vfsmounts  
until it does an absolute chdir().  Shells tend not to be terribly  
useful for this because they use absolute paths more than necessary,  
but I think you can make it work.  For instance, if you run these  
commands after mounting the root filesystem:

mount -t proc proc /realroot/proc
cd /
mount --move /realroot /
find /proc/$$/cwd/ -mindepth 1 -maxdepth 1 -print0 \
		| xargs -0 rm -rf -- '{}' ';'
cd /

This will set up proc on the new root filesystem, move /realroot to  
the root directory, and delete the contents of the old initramfs  
using the old current-working-directory link of your shell.  The  
amusing thing is that you are using the "find" and "rm" commands from  
the newly mounted root filesystem to remove the contents of the  
completely hidden initramfs.  The only reason it's actually possible  
is because you have the tenuous thread of the working directory.

Cheers,
Kyle Moffett

Re: pivot_root alternative

[Salvatore De Paolis]

So i choice to use switch_root which is already inside the busybox but with a
bit(?) of unluck cause it still wont run.

It hangs on !S_ISREG(st1.st_mode)

Can you explain to me how to interpret it?

Regards
S.