Re: pivot_root solved by patch to 2.4.22-pre7

Re: pivot_root solved by patch to 2.4.22-pre7

[Rene Mayrhofer]

Hi all,

The problem with pivot_root that appeared in 2.4.21-ac4 and the 
2.4.22-pre kernels is now solved (at least for my case) by applying the 
trvial patch sent by Jason Baron.

Jason Baron wrote:
> right. so the semantics of how file tables are shared has changed a bit. I
> would think that for at least 'init', it'd be nice to preserve the
> original behavior, for situations such as you described. Something like
> the following would probably work, although i havent' tried the test
> script.
> 
> --- linux/kernel/fork.c.orig  2003-07-23 21:34:59.000000000 -0400
> +++ linux/kernel/fork.c       2003-07-23 21:35:45.000000000 -0400
> @@ -558,7 +558,7 @@ int unshare_files(void)
>  
>         /* This can race but the race causes us to copy when we don't
>            need to and drop the copy */
> -       if(atomic_read(&files->count) == 1)
> +       if(atomic_read(&files->count) == 1 || (current->pid == 1))
>         {
>                 atomic_inc(&files->count);
>                 return 0;
>    



I tried that on my system and it works as expected. The kernel processes
close their fds and the old root fs can thus be unmounted after
pivot_root. Thanks for the hint !
So the problem is solved for me and it would be wonderful to get it into
2.4.22.

best regards,
Rene

Re: pivot_root solved by patch to 2.4.22-pre7

[Marcelo Tosatti]

What is your problem with pivot_root? 

Sorry but I've searched the list archives and found nothing but this email 
and this patch, which seems a bit hackish.

On Wed, 6 Aug 2003, Rene Mayrhofer wrote:

> Hi all,
> 
> The problem with pivot_root that appeared in 2.4.21-ac4 and the 
> 2.4.22-pre kernels is now solved (at least for my case) by applying the 
> trvial patch sent by Jason Baron.
> 
> Jason Baron wrote:
> > right. so the semantics of how file tables are shared has changed a bit. I
> > would think that for at least 'init', it'd be nice to preserve the
> > original behavior, for situations such as you described. Something like
> > the following would probably work, although i havent' tried the test
> > script.
> > 
> > --- linux/kernel/fork.c.orig  2003-07-23 21:34:59.000000000 -0400
> > +++ linux/kernel/fork.c       2003-07-23 21:35:45.000000000 -0400
> > @@ -558,7 +558,7 @@ int unshare_files(void)
> >  
> >         /* This can race but the race causes us to copy when we don't
> >            need to and drop the copy */
> > -       if(atomic_read(&files->count) == 1)
> > +       if(atomic_read(&files->count) == 1 || (current->pid == 1))
> >         {
> >                 atomic_inc(&files->count);
> >                 return 0;
> >    
> 
> 
> 
> I tried that on my system and it works as expected. The kernel processes
> close their fds and the old root fs can thus be unmounted after
> pivot_root. Thanks for the hint !
> So the problem is solved for me and it would be wonderful to get it into
> 2.4.22.
> 
> best regards,
> Rene
> 
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
> 

Re: pivot_root solved by patch to 2.4.22-pre7

[Herbert Pötzl]

On Wed, Aug 06, 2003 at 04:33:52PM -0300, Marcelo Tosatti wrote:
> 
> What is your problem with pivot_root? 
> 
> Sorry but I've searched the list archives and found nothing but this email 
> and this patch, which seems a bit hackish.

 Jul 21  Rene Mayrhofer     82 pivot_root seems to be broken in 2.4.21-a1716
 Jul 22  Denis Vlasenko    103   >
 Jul 22  Rene Mayrhofer     54     >
 Jul 22  Jason Baron        18       >
 Jul 22  Alan Cox           22         >
 Jul 22  Rene Mayrhofer     25           >
 Jul 22  Alan Cox           17             >
 Jul 22  Rene Mayrhofer    154               >
 Jul 23  Mika Penttilä      40               >
 Jul 22  Mika Penttilä      41           >

HTH,
Herbert

> On Wed, 6 Aug 2003, Rene Mayrhofer wrote:
> 
> > Hi all,
> > 
> > The problem with pivot_root that appeared in 2.4.21-ac4 and the 
> > 2.4.22-pre kernels is now solved (at least for my case) by applying the 
> > trvial patch sent by Jason Baron.
> > 
> > Jason Baron wrote:
> > > right. so the semantics of how file tables are shared has changed a bit. I
> > > would think that for at least 'init', it'd be nice to preserve the
> > > original behavior, for situations such as you described. Something like
> > > the following would probably work, although i havent' tried the test
> > > script.
> > > 
> > > --- linux/kernel/fork.c.orig  2003-07-23 21:34:59.000000000 -0400
> > > +++ linux/kernel/fork.c       2003-07-23 21:35:45.000000000 -0400
> > > @@ -558,7 +558,7 @@ int unshare_files(void)
> > >  
> > >         /* This can race but the race causes us to copy when we don't
> > >            need to and drop the copy */
> > > -       if(atomic_read(&files->count) == 1)
> > > +       if(atomic_read(&files->count) == 1 || (current->pid == 1))
> > >         {
> > >                 atomic_inc(&files->count);
> > >                 return 0;
> > >    
> > 
> > 
> > 
> > I tried that on my system and it works as expected. The kernel processes
> > close their fds and the old root fs can thus be unmounted after
> > pivot_root. Thanks for the hint !
> > So the problem is solved for me and it would be wonderful to get it into
> > 2.4.22.
> > 
> > best regards,
> > Rene
> > 
> > -
> > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> > the body of a message to majordomo@vger.kernel.org
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> > Please read the FAQ at  http://www.tux.org/lkml/
> > 
> 
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

Re: pivot_root solved by patch to 2.4.22-pre7

[Andreas Haumer]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi!

Herbert Pötzl wrote:
> On Wed, Aug 06, 2003 at 04:33:52PM -0300, Marcelo Tosatti wrote:
>
>>What is your problem with pivot_root?
>>
>>Sorry but I've searched the list archives and found nothing but this email
>>and this patch, which seems a bit hackish.
>
>
>  Jul 21  Rene Mayrhofer     82 pivot_root seems to be broken in 2.4.21-a1716
>  Jul 22  Denis Vlasenko    103   >
>  Jul 22  Rene Mayrhofer     54     >
>  Jul 22  Jason Baron        18       >
>  Jul 22  Alan Cox           22         >
>  Jul 22  Rene Mayrhofer     25           >
>  Jul 22  Alan Cox           17             >
>  Jul 22  Rene Mayrhofer    154               >
>  Jul 23  Mika Penttilä      40               >
>  Jul 22  Mika Penttilä      41           >
>
This patch also solves my problem as reported on Jul. 29th
(Subject: "2.4.22-pre4: devfs on initrd stays busy after pivot_root"
Message-ID: 3F267FD7.4040400@xss.co.at) and in more detail
on July 30th, (Message-ID: 3F2795DE.5020306@xss.co.at)

I just tried it with 2.4.22-rc1, and with this patch I am
able to umount /initrd/dev and /initrd after pivot_root
again!

Don't know yet if it has any ill side effects, though.

HTH

- - andreas

- --
Andreas Haumer                     | mailto:andreas@xss.co.at
*x Software + Systeme              | http://www.xss.co.at/
Karmarschgasse 51/2/20             | Tel: +43-1-6060114-0
A-1100 Vienna, Austria             | Fax: +43-1-6060114-71
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.1 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQE/Mh0/xJmyeGcXPhERAmbLAKCUrXFUA44ggJvOBMlwvA4GKFsWdwCeNIid
4bWSyQ2OFqwXnNdU8t/bJsw=
=sdiS
-----END PGP SIGNATURE-----

Re: pivot_root solved by patch to 2.4.22-pre7

[Rene Mayrhofer]

Andreas Haumer wrote:
> I just tried it with 2.4.22-rc1, and with this patch I am
> able to umount /initrd/dev and /initrd after pivot_root
> again!
Same for me. With this patch, all the 2.4.21-ac*, 2.4.22-pre*-ac* and 
2.4.22-rc1 kernels I tried work again.

> Don't know yet if it has any ill side effects, though.
Doesn't seem to have, but IANAKG (I am not a kernel guru) ;)

- Rene

Re: pivot_root solved by patch to 2.4.22-pre7

[Alan Cox]

On Mer, 2003-08-06 at 07:27, Rene Mayrhofer wrote:
> Hi all,
> 
> The problem with pivot_root that appeared in 2.4.21-ac4 and the 
> 2.4.22-pre kernels is now solved (at least for my case) by applying the 
> trvial patch sent by Jason Baron.

The patch shouldnt be needed or make any difference. I have to
understand why its fixing the problem and fix it properly yet (or
someone does)

Re: pivot_root solved by patch to 2.4.22-pre7

[Jason Baron]

On 7 Aug 2003, Alan Cox wrote:

> On Mer, 2003-08-06 at 07:27, Rene Mayrhofer wrote:
> > Hi all,
> > 
> > The problem with pivot_root that appeared in 2.4.21-ac4 and the 
> > 2.4.22-pre kernels is now solved (at least for my case) by applying the 
> > trvial patch sent by Jason Baron.
> 
> The patch shouldnt be needed or make any difference. I have to
> understand why its fixing the problem and fix it properly yet (or
> someone does)
> 
> 
 
it clearly makes a difference.

the unshare_files change causes init to no longer share the same fd table
with the other kernel threads. thus, when init closes or opens fds it does
not affect the other kernel threads. this patch allows init to continue to
share the same fd table with the other kernel threads as before the
unshare_files change. it does not compromise the intention of the
unshare_files change afaik. 

-Jason

Re: pivot_root solved by patch to 2.4.22-pre7

[Alan Cox]

On Iau, 2003-08-07 at 16:26, Jason Baron wrote:
> it clearly makes a difference.
> 
> the unshare_files change causes init to no longer share the same fd table
> with the other kernel threads. thus, when init closes or opens fds it does

Ah yes.. because of do_basic_setup. Having /sbin/init sharing with
kernel threads doesn't actually strike me as too clever anyway although
none of them should be using fd stuff.

In which case I guess we should call unshare_files directly before we
open /dev/console in init/main.c.

Re: pivot_root solved by patch to 2.4.22-pre7

[Jason Baron]

On 7 Aug 2003, Alan Cox wrote:

> On Iau, 2003-08-07 at 16:26, Jason Baron wrote:
> > it clearly makes a difference.
> > 
> > the unshare_files change causes init to no longer share the same fd table
> > with the other kernel threads. thus, when init closes or opens fds it does
> 
> Ah yes.. because of do_basic_setup. Having /sbin/init sharing with
> kernel threads doesn't actually strike me as too clever anyway although
> none of them should be using fd stuff.
> 
> In which case I guess we should call unshare_files directly before we
> open /dev/console in init/main.c.
> 

ok, but what if we re-exec init a couple of times? 

Re: pivot_root solved by patch to 2.4.22-pre7

[Jason Lunz]

In gmane.linux.kernel, you wrote:
>> the unshare_files change causes init to no longer share the same fd table
>> with the other kernel threads. thus, when init closes or opens fds it does
> 
> Ah yes.. because of do_basic_setup. Having /sbin/init sharing with
> kernel threads doesn't actually strike me as too clever anyway although
> none of them should be using fd stuff.
> 
> In which case I guess we should call unshare_files directly before we
> open /dev/console in init/main.c.

Is this going to be fixed for 2.4.22? In -rc2, I still get this after
pivot_root (I'm using pivot_root, but not initrd):

	halfoat:0:~ # umount /mnt
	umount: /mnt: device is busy
	halfoat:1:~ # lsof /mnt
	COMMAND   PID USER   FD   TYPE DEVICE SIZE NODE NAME
	keventd     2 root    0u   CHR    5,1        21 /mnt/dev/console
	keventd     2 root    1u   CHR    5,1        21 /mnt/dev/console
	keventd     2 root    2u   CHR    5,1        21 /mnt/dev/console
	ksoftirqd   3 root    0u   CHR    5,1        21 /mnt/dev/console
	ksoftirqd   3 root    1u   CHR    5,1        21 /mnt/dev/console
	ksoftirqd   3 root    2u   CHR    5,1        21 /mnt/dev/console
	ksoftirqd   4 root    0u   CHR    5,1        21 /mnt/dev/console
	ksoftirqd   4 root    1u   CHR    5,1        21 /mnt/dev/console
	ksoftirqd   4 root    2u   CHR    5,1        21 /mnt/dev/console
	kswapd      5 root    0u   CHR    5,1        21 /mnt/dev/console
	kswapd      5 root    1u   CHR    5,1        21 /mnt/dev/console
	kswapd      5 root    2u   CHR    5,1        21 /mnt/dev/console
	bdflush     6 root    0u   CHR    5,1        21 /mnt/dev/console
	bdflush     6 root    1u   CHR    5,1        21 /mnt/dev/console
	bdflush     6 root    2u   CHR    5,1        21 /mnt/dev/console
	kupdated    7 root    0u   CHR    5,1        21 /mnt/dev/console
	kupdated    7 root    1u   CHR    5,1        21 /mnt/dev/console
	kupdated    7 root    2u   CHR    5,1        21 /mnt/dev/console
	kjournald  64 root    0u   CHR    5,1        21 /mnt/dev/console
	kjournald  64 root    1u   CHR    5,1        21 /mnt/dev/console
	kjournald  64 root    2u   CHR    5,1        21 /mnt/dev/console

Jason

Re: pivot_root solved by patch to 2.4.22-pre7

[Mathias Sundman]

> > Ah yes.. because of do_basic_setup. Having /sbin/init sharing with
> > kernel threads doesn't actually strike me as too clever anyway 
although
> > none of them should be using fd stuff.
> > 
> > In which case I guess we should call unshare_files directly before we
> > open /dev/console in init/main.c.
> 
> Is this going to be fixed for 2.4.22? In -rc2, I still get this after
> pivot_root (I'm using pivot_root, but not initrd):
> 
>                halfoat:0:~ # umount /mnt
>                umount: /mnt: device is busy

Have this problem been resolved yet, or do one
still need to use the patch supplied by Jason Baron?

/Mathias