Early image startup

Early image startup

[Gary Thomas]

I have a Poky image which uses a number of kernel modules.  For
whatever reason, the raw image has not had the modules configured
(no /lib/modules/.../modules.dep)  This file is created by 'update-modules'
but that is run too late in the boot process.  The problem is that
other code, e.g. I'm using iptables via modules, requires the modules
be configured early and this fails.  If I reboot (after 'update-modules'
has run), everything is fine.

My system is PowerPC and I'm using ipk packaging.

Here's a snippet from my initial boot log:
   Please wait: booting...
   Starting udev
   Configuring network interfaces... udhcpc (v1.17.3) started
   Setting IP address 0.0.0.0 on eth0
   Sending discover...
   PHY: mdio@ff024520:01 - Link is Up - 1000/Full
   Sending discover...
     Sending select for 192.168.12.140...
   Lease of 192.168.12.140 obtained, lease time 600
   Setting IP address 192.168.12.140 on eth0
   Deleting routers
   Adding router 192.168.12.1
   Recreating /etc/resolv.conf
    Adding DNS server 192.168.12.1
   done.
   modprobe: can't open 'modules.dep': No such file or directory
   iptables v1.4.9: can't initialize iptables table `filter': iptables who? (do you need to insmod?)
   Perhaps iptables or your kernel needs to be upgraded.
   Starting DHCP server: .
   Fri Apr  8 12:29:00 UTC 2011
   Configuring update-modules.
   Configuring kernel-image-2.6.37.3.
   Configuring kernel-module-nf-conntrack.
   Configuring kernel-module-x-tables.
   Configuring kernel-module-ip-tables.
   Configuring kernel-module-nf-defrag-ipv4.
   Configuring kernel-module-nf-conntrack-ipv4.
   Configuring kernel-module-iptable-mangle.
   Configuring kernel-module-xt-state.
   Configuring kernel-module-nf-nat.
   Configuring kernel-module-nf-conntrack-tftp.
   Configuring kernel-module-nf-nat-tftp.
   Configuring kernel-module-ipt-masquerade.
   Configuring kernel-module-iptable-nat.
   Configuring kernel-module-iptable-filter.
   Configuring netkit-tftpd.
   Configuring kernel-module-xt-tcpudp.

So, what decides when these 'configuring' steps run?

Thanks

-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------

Re: Early image startup

[Gary Thomas]

On 04/08/2011 07:20 AM, Gary Thomas wrote:
> I have a Poky image which uses a number of kernel modules. For
> whatever reason, the raw image has not had the modules configured
> (no /lib/modules/.../modules.dep) This file is created by 'update-modules'
> but that is run too late in the boot process. The problem is that
> other code, e.g. I'm using iptables via modules, requires the modules
> be configured early and this fails. If I reboot (after 'update-modules'
> has run), everything is fine.
>
> My system is PowerPC and I'm using ipk packaging.
>
> Here's a snippet from my initial boot log:
> Please wait: booting...
> Starting udev
> Configuring network interfaces... udhcpc (v1.17.3) started
> Setting IP address 0.0.0.0 on eth0
> Sending discover...
> PHY: mdio@ff024520:01 - Link is Up - 1000/Full
> Sending discover...
> Sending select for 192.168.12.140...
> Lease of 192.168.12.140 obtained, lease time 600
> Setting IP address 192.168.12.140 on eth0
> Deleting routers
> Adding router 192.168.12.1
> Recreating /etc/resolv.conf
> Adding DNS server 192.168.12.1
> done.
> modprobe: can't open 'modules.dep': No such file or directory
> iptables v1.4.9: can't initialize iptables table `filter': iptables who? (do you need to insmod?)
> Perhaps iptables or your kernel needs to be upgraded.
> Starting DHCP server: .
> Fri Apr 8 12:29:00 UTC 2011
> Configuring update-modules.
> Configuring kernel-image-2.6.37.3.
> Configuring kernel-module-nf-conntrack.
> Configuring kernel-module-x-tables.
> Configuring kernel-module-ip-tables.
> Configuring kernel-module-nf-defrag-ipv4.
> Configuring kernel-module-nf-conntrack-ipv4.
> Configuring kernel-module-iptable-mangle.
> Configuring kernel-module-xt-state.
> Configuring kernel-module-nf-nat.
> Configuring kernel-module-nf-conntrack-tftp.
> Configuring kernel-module-nf-nat-tftp.
> Configuring kernel-module-ipt-masquerade.
> Configuring kernel-module-iptable-nat.
> Configuring kernel-module-iptable-filter.
> Configuring netkit-tftpd.
> Configuring kernel-module-xt-tcpudp.
>
> So, what decides when these 'configuring' steps run?

FAOD, these are my init.S steps:
   /etc/rcS.d:
    S02banner.sh -> ../init.d/banner.sh
    S03sysfs.sh -> ../init.d/sysfs.sh
    S04udev -> ../init.d/udev
    S10checkroot.sh -> ../init.d/checkroot.sh
    S20modutils.sh -> ../init.d/modutils.sh
    S35mountall.sh -> ../init.d/mountall.sh
    S36udev-cache -> ../init.d/udev-cache
    S37populate-volatile.sh -> ../init.d/populate-volatile.sh
    S38devpts.sh -> ../init.d/devpts.sh
    S39hostname.sh -> ../init.d/hostname.sh
    S40networking -> ../init.d/networking
    S41nat_setup -> ../init.d/nat_setup
    S45mountnfs.sh -> ../init.d/mountnfs.sh
    S50dhcp-server -> ../init.d/dhcp-server
    S55bootmisc.sh -> ../init.d/bootmisc.sh


-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------

Re: [poky] Early image startup

[Koen Kooi]

Op 8 apr 2011, om 15:20 heeft Gary Thomas het volgende geschreven:

> I have a Poky image which uses a number of kernel modules.  For
> whatever reason, the raw image has not had the modules configured
> (no /lib/modules/.../modules.dep)  This file is created by 'update-modules'
> but that is run too late in the boot process.  The problem is that
> other code, e.g. I'm using iptables via modules, requires the modules
> be configured early and this fails.  If I reboot (after 'update-modules'
> has run), everything is fine.

Sound like you need http://patches.openembedded.org/patch/1917/

Re: [poky] Early image startup

[Richard Purdie]

On Fri, 2011-04-08 at 15:30 +0200, Koen Kooi wrote:
> Op 8 apr 2011, om 15:20 heeft Gary Thomas het volgende geschreven:
> 
> > I have a Poky image which uses a number of kernel modules.  For
> > whatever reason, the raw image has not had the modules configured
> > (no /lib/modules/.../modules.dep)  This file is created by 'update-modules'
> > but that is run too late in the boot process.  The problem is that
> > other code, e.g. I'm using iptables via modules, requires the modules
> > be configured early and this fails.  If I reboot (after 'update-modules'
> > has run), everything is fine.
> 
> Sound like you need http://patches.openembedded.org/patch/1917/

To be honest that sounds like a more fundamental problem as if the user
cares which rtc should be rtc0, there are other ordering constraints
being imposed such as a symlink to the "best" rtc device...

Cheers,

Richard

Re: [poky] Early image startup

[Gary Thomas]

On 04/08/2011 07:50 AM, Richard Purdie wrote:
> On Fri, 2011-04-08 at 15:30 +0200, Koen Kooi wrote:
>> Op 8 apr 2011, om 15:20 heeft Gary Thomas het volgende geschreven:
>>
>>> I have a Poky image which uses a number of kernel modules.  For
>>> whatever reason, the raw image has not had the modules configured
>>> (no /lib/modules/.../modules.dep)  This file is created by 'update-modules'
>>> but that is run too late in the boot process.  The problem is that
>>> other code, e.g. I'm using iptables via modules, requires the modules
>>> be configured early and this fails.  If I reboot (after 'update-modules'
>>> has run), everything is fine.
>>
>> Sound like you need http://patches.openembedded.org/patch/1917/
>
> To be honest that sounds like a more fundamental problem as if the user
> cares which rtc should be rtc0, there are other ordering constraints
> being imposed such as a symlink to the "best" rtc device...

Indeed, this patch did not help my situation at all.

I think my problem is that of post-install package scripts (hence
'Configuring ...') not being run until quite late.

-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------

Re: [poky] Early image startup

[Gary Thomas]

On 04/08/2011 08:01 AM, Gary Thomas wrote:
> On 04/08/2011 07:50 AM, Richard Purdie wrote:
>> On Fri, 2011-04-08 at 15:30 +0200, Koen Kooi wrote:
>>> Op 8 apr 2011, om 15:20 heeft Gary Thomas het volgende geschreven:
>>>
>>>> I have a Poky image which uses a number of kernel modules. For
>>>> whatever reason, the raw image has not had the modules configured
>>>> (no /lib/modules/.../modules.dep) This file is created by 'update-modules'
>>>> but that is run too late in the boot process. The problem is that
>>>> other code, e.g. I'm using iptables via modules, requires the modules
>>>> be configured early and this fails. If I reboot (after 'update-modules'
>>>> has run), everything is fine.
>>>
>>> Sound like you need http://patches.openembedded.org/patch/1917/
>>
>> To be honest that sounds like a more fundamental problem as if the user
>> cares which rtc should be rtc0, there are other ordering constraints
>> being imposed such as a symlink to the "best" rtc device...
>
> Indeed, this patch did not help my situation at all.
>
> I think my problem is that of post-install package scripts (hence
> 'Configuring ...') not being run until quite late.

Looking at the pre-run image, I see that there is /etc/rcS.d/S98configure
which runs these scripts.  Is there any reason I can't move it earlier?

-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------

Re: [poky] Early image startup

[Gary Thomas]

On 04/08/2011 08:07 AM, Gary Thomas wrote:
> On 04/08/2011 08:01 AM, Gary Thomas wrote:
>> On 04/08/2011 07:50 AM, Richard Purdie wrote:
>>> On Fri, 2011-04-08 at 15:30 +0200, Koen Kooi wrote:
>>>> Op 8 apr 2011, om 15:20 heeft Gary Thomas het volgende geschreven:
>>>>
>>>>> I have a Poky image which uses a number of kernel modules. For
>>>>> whatever reason, the raw image has not had the modules configured
>>>>> (no /lib/modules/.../modules.dep) This file is created by 'update-modules'
>>>>> but that is run too late in the boot process. The problem is that
>>>>> other code, e.g. I'm using iptables via modules, requires the modules
>>>>> be configured early and this fails. If I reboot (after 'update-modules'
>>>>> has run), everything is fine.
>>>>
>>>> Sound like you need http://patches.openembedded.org/patch/1917/
>>>
>>> To be honest that sounds like a more fundamental problem as if the user
>>> cares which rtc should be rtc0, there are other ordering constraints
>>> being imposed such as a symlink to the "best" rtc device...
>>
>> Indeed, this patch did not help my situation at all.
>>
>> I think my problem is that of post-install package scripts (hence
>> 'Configuring ...') not being run until quite late.
>
> Looking at the pre-run image, I see that there is /etc/rcS.d/S98configure
> which runs these scripts. Is there any reason I can't move it earlier?

Found it:
   ipk systems - meta/recipes-devtools/opkg/opkg_0.1.8.bb
   rpm systems - meta/classes/rootfs_rpm.bbclass

Note: the ipk version is quasi-configurable, the rpm version hardwired at 98.
Since these are package install scripts, which presumable could have been run
at image creation time but needed to run on the target hardware, why put it
off to so late?  I would think that this should be done _first_, not _last_

-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------

Re: [poky] Early image startup

[Hatle, Mark]

As for the rpm ordering this was preserved from an earlier version.  If there is a more appropriate place to run the late configuration scripts, let's move it.  I think that it's likely a good place to move it to soon after the filesystem has been mounted RW.  Do any of the scripts normally require networking or other system resources?



On Apr 8, 2011, at 7:18 AM, "Gary Thomas" <gary@mlbassoc.com> wrote:

> On 04/08/2011 08:07 AM, Gary Thomas wrote:
>> On 04/08/2011 08:01 AM, Gary Thomas wrote:
>>> On 04/08/2011 07:50 AM, Richard Purdie wrote:
>>>> On Fri, 2011-04-08 at 15:30 +0200, Koen Kooi wrote:
>>>>> Op 8 apr 2011, om 15:20 heeft Gary Thomas het volgende geschreven:
>>>>> 
>>>>>> I have a Poky image which uses a number of kernel modules. For
>>>>>> whatever reason, the raw image has not had the modules configured
>>>>>> (no /lib/modules/.../modules.dep) This file is created by 'update-modules'
>>>>>> but that is run too late in the boot process. The problem is that
>>>>>> other code, e.g. I'm using iptables via modules, requires the modules
>>>>>> be configured early and this fails. If I reboot (after 'update-modules'
>>>>>> has run), everything is fine.
>>>>> 
>>>>> Sound like you need http://patches.openembedded.org/patch/1917/
>>>> 
>>>> To be honest that sounds like a more fundamental problem as if the user
>>>> cares which rtc should be rtc0, there are other ordering constraints
>>>> being imposed such as a symlink to the "best" rtc device...
>>> 
>>> Indeed, this patch did not help my situation at all.
>>> 
>>> I think my problem is that of post-install package scripts (hence
>>> 'Configuring ...') not being run until quite late.
>> 
>> Looking at the pre-run image, I see that there is /etc/rcS.d/S98configure
>> which runs these scripts. Is there any reason I can't move it earlier?
> 
> Found it:
>  ipk systems - meta/recipes-devtools/opkg/opkg_0.1.8.bb
>  rpm systems - meta/classes/rootfs_rpm.bbclass
> 
> Note: the ipk version is quasi-configurable, the rpm version hardwired at 98.
> Since these are package install scripts, which presumable could have been run
> at image creation time but needed to run on the target hardware, why put it
> off to so late?  I would think that this should be done _first_, not _last_
> 
> -- 
> ------------------------------------------------------------
> Gary Thomas                 |  Consulting for the
> MLB Associates              |    Embedded world
> ------------------------------------------------------------
> 
> _______________________________________________
> Openembedded-core mailing list
> Openembedded-core@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-core

Re: [poky] Early image startup

[Gary Thomas]

On 04/08/2011 10:40 AM, Hatle, Mark wrote:
> As for the rpm ordering this was preserved from an earlier version.  If there is a more appropriate place to run the late configuration scripts, let's move it.  I think that it's likely a good place to move it to soon after the filesystem has been mounted RW.  Do any of the scripts normally require networking or other system resources?

Hard to say what's perfect, but I certainly don't believe that they need
to be put off to the very last.   In any case, I think the best thing is
to make it easily adjustable, per my patch.

Can we please keep this discussion on the Poky list where it started?
It seems that OE (and maybe OE-core) handle this differently - I've not
looked in depth - and I'm mostly interested in getting my Poky builds
healthy.

> On Apr 8, 2011, at 7:18 AM, "Gary Thomas"<gary@mlbassoc.com>  wrote:
>
>> On 04/08/2011 08:07 AM, Gary Thomas wrote:
>>> On 04/08/2011 08:01 AM, Gary Thomas wrote:
>>>> On 04/08/2011 07:50 AM, Richard Purdie wrote:
>>>>> On Fri, 2011-04-08 at 15:30 +0200, Koen Kooi wrote:
>>>>>> Op 8 apr 2011, om 15:20 heeft Gary Thomas het volgende geschreven:
>>>>>>
>>>>>>> I have a Poky image which uses a number of kernel modules. For
>>>>>>> whatever reason, the raw image has not had the modules configured
>>>>>>> (no /lib/modules/.../modules.dep) This file is created by 'update-modules'
>>>>>>> but that is run too late in the boot process. The problem is that
>>>>>>> other code, e.g. I'm using iptables via modules, requires the modules
>>>>>>> be configured early and this fails. If I reboot (after 'update-modules'
>>>>>>> has run), everything is fine.
>>>>>>
>>>>>> Sound like you need http://patches.openembedded.org/patch/1917/
>>>>>
>>>>> To be honest that sounds like a more fundamental problem as if the user
>>>>> cares which rtc should be rtc0, there are other ordering constraints
>>>>> being imposed such as a symlink to the "best" rtc device...
>>>>
>>>> Indeed, this patch did not help my situation at all.
>>>>
>>>> I think my problem is that of post-install package scripts (hence
>>>> 'Configuring ...') not being run until quite late.
>>>
>>> Looking at the pre-run image, I see that there is /etc/rcS.d/S98configure
>>> which runs these scripts. Is there any reason I can't move it earlier?
>>
>> Found it:
>>   ipk systems - meta/recipes-devtools/opkg/opkg_0.1.8.bb
>>   rpm systems - meta/classes/rootfs_rpm.bbclass
>>
>> Note: the ipk version is quasi-configurable, the rpm version hardwired at 98.
>> Since these are package install scripts, which presumable could have been run
>> at image creation time but needed to run on the target hardware, why put it
>> off to so late?  I would think that this should be done _first_, not _last_
>>
>> --
>> ------------------------------------------------------------
>> Gary Thomas                 |  Consulting for the
>> MLB Associates              |    Embedded world
>> ------------------------------------------------------------
>>
>> _______________________________________________
>> Openembedded-core mailing list
>> Openembedded-core@lists.openembedded.org
>> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-core
>
> _______________________________________________
> Openembedded-core mailing list
> Openembedded-core@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-core

-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------

Re: [OE-core] Early image startup

[Gary Thomas]

On 04/08/2011 10:40 AM, Hatle, Mark wrote:
> As for the rpm ordering this was preserved from an earlier version.  If there is a more appropriate place to run the late configuration scripts, let's move it.  I think that it's likely a good place to move it to soon after the filesystem has been mounted RW.  Do any of the scripts normally require networking or other system resources?

Hard to say what's perfect, but I certainly don't believe that they need
to be put off to the very last.   In any case, I think the best thing is
to make it easily adjustable, per my patch.

Can we please keep this discussion on the Poky list where it started?
It seems that OE (and maybe OE-core) handle this differently - I've not
looked in depth - and I'm mostly interested in getting my Poky builds
healthy.

> On Apr 8, 2011, at 7:18 AM, "Gary Thomas"<gary@mlbassoc.com>  wrote:
>
>> On 04/08/2011 08:07 AM, Gary Thomas wrote:
>>> On 04/08/2011 08:01 AM, Gary Thomas wrote:
>>>> On 04/08/2011 07:50 AM, Richard Purdie wrote:
>>>>> On Fri, 2011-04-08 at 15:30 +0200, Koen Kooi wrote:
>>>>>> Op 8 apr 2011, om 15:20 heeft Gary Thomas het volgende geschreven:
>>>>>>
>>>>>>> I have a Poky image which uses a number of kernel modules. For
>>>>>>> whatever reason, the raw image has not had the modules configured
>>>>>>> (no /lib/modules/.../modules.dep) This file is created by 'update-modules'
>>>>>>> but that is run too late in the boot process. The problem is that
>>>>>>> other code, e.g. I'm using iptables via modules, requires the modules
>>>>>>> be configured early and this fails. If I reboot (after 'update-modules'
>>>>>>> has run), everything is fine.
>>>>>>
>>>>>> Sound like you need http://patches.openembedded.org/patch/1917/
>>>>>
>>>>> To be honest that sounds like a more fundamental problem as if the user
>>>>> cares which rtc should be rtc0, there are other ordering constraints
>>>>> being imposed such as a symlink to the "best" rtc device...
>>>>
>>>> Indeed, this patch did not help my situation at all.
>>>>
>>>> I think my problem is that of post-install package scripts (hence
>>>> 'Configuring ...') not being run until quite late.
>>>
>>> Looking at the pre-run image, I see that there is /etc/rcS.d/S98configure
>>> which runs these scripts. Is there any reason I can't move it earlier?
>>
>> Found it:
>>   ipk systems - meta/recipes-devtools/opkg/opkg_0.1.8.bb
>>   rpm systems - meta/classes/rootfs_rpm.bbclass
>>
>> Note: the ipk version is quasi-configurable, the rpm version hardwired at 98.
>> Since these are package install scripts, which presumable could have been run
>> at image creation time but needed to run on the target hardware, why put it
>> off to so late?  I would think that this should be done _first_, not _last_
>>
>> --
>> ------------------------------------------------------------
>> Gary Thomas                 |  Consulting for the
>> MLB Associates              |    Embedded world
>> ------------------------------------------------------------
>>
>> _______________________________________________
>> Openembedded-core mailing list
>> Openembedded-core@lists.openembedded.org
>> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-core
>
> _______________________________________________
> Openembedded-core mailing list
> Openembedded-core@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-core

-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------