Missconfiguration of ethernet interfaces on reboot

Hello everybody,

recently I have the same error in two different machines running
RedHat Linux 9 (kernel 2.4.20-8).

Each machine has 3 NICs: 2 of them come included in motherboard (one
network card with 2 NICs: one Intel e100 and one intel e1000) and the
other is an Intel e1000 Fiber Network Card attached in a PCI-Express
slot.

The interfaces' alias are assigned like this (looking at /etc/
modules.conf):
eth0: e100 NIC in motherboard
eth1: e1000 NIC in mother board
eth2: e1000 NIC in PCI slot

All the system was working properly untill the other day the machine
made a reboot (due to the softdog) and the interfaces' configuration
became wrong: the system couldn't find the eth2 interface and also it
tried to asign eth0 a e1000 module, but it couldn't, so it didn't
result in a good rising of the interface and failed to be up.
Indeed, the e100 module was not loaded (consulting via lsmod)

When I tried to re-configure the interfaces trough the redhat-config-
networg assistant, then appeared a message box telling me that I was
wrongly trying to asign a e100 module to eth0 while it needed a e1000
module (it's false).

It seems like someone is saying that interface eth0 needs a e1000
module, but neither the /etc/module.conf file, nor the /etc/sysconfig/
network-scripts/ifcfg-ethX files had changed.

I looked at the /etc/sysconfig/hwconf file and it was the same like
few weeks ago. No changes. But I realized that on the interface name
especification, the COMPLETE interface name was not correct.
I mean, it was written just eth, without 0, 1 or 2, insteed of
eth0,eth1,eth2.

The file hwconf network part is like this:

class: NETWORK
bus: PCI
detached: 0
device: eth
driver: e100
desc: "Intel Corp.|82801BD PRO/100 VE (LOM) Ethernet Controller"
vendorId: 8086
deviceId: 1039
subVendorId: 8086
subDeviceId: 103a
pciType: 1
-
class: NETWORK
bus: PCI
detached: 0
device: eth
driver: e1000
desc: "Unknown vendor|Generic e1000 device"
vendorId: 8086
deviceId: 1076
subVendorId: 8086
subDeviceId: 1076
pciType: 1
-
class: NETWORK
bus: PCI
detached: 0
device: eth
driver: e1000
desc: "Unknown vendor|Generic e1000 device"
vendorId: 8086
deviceId: 1027
subVendorId: 8086
subDeviceId: 1027
pciType: 1


In other (newer) RedHat systems (kernel 2.6.9-55), interface name in
that file is complete (eth0, eth1, eth2).


The /etc/modules.conf file is like this:
alias eth0 e100
alias eth1 e1000
alias eth2 e1000


Does anybody knows if there's a bug in hwconf file or something
related with those files? I was several days looking for a clue about
that, but I haven't find anything.


Thanks for any advice,
Fionn

On Mon, 26 May 2008, in the Usenet newsgroup comp.os.linux.networking, in
article <52633167-2a60-4a3c-89c4-e80c4c64f04d@f36g2000hsa.googlegroups.com>,
Fionn wrote:

NOTE: Posting from groups.google.com (or some web-forums) dramatically
reduces the chance of your post being seen. Find a real news server.

>recently I have the same error in two different machines running
>RedHat Linux 9 (kernel 2.4.20-8).

That's the original kernel on an unmaintained 5 year old system.
There were at least 9 kernel errata during the supported life, and
three more backports - ending with 2.4.20-46.9.legacy in March 2006.

>All the system was working properly untill the other day the machine
>made a reboot (due to the softdog)

Both systems suffered the same fault at the same time???

>and the interfaces' configuration became wrong: the system couldn't
>find the eth2 interface and also it tried to asign eth0 a e1000
>module, but it couldn't, so it didn't result in a good rising of the
>interface and failed to be up. Indeed, the e100 module was not loaded
>(consulting via lsmod)

That's usually a hardware or BIOS problem, where the e100 NIC isn't
being found for some reason. Look at the boot messages in
/var/log/messages relating to finding the cards. RH9 had a piece of
crap application called 'kudzu' that was meant to reconfigure the
system when hardware changed. I usually uninstalled that package as
my hardware wasn't being changed every time the systems reboot.

>driver: e1000
>desc: "Unknown vendor|Generic e1000 device"
>vendorId: 8086
>deviceId: 1076

That's another indication that the kernel is obsolete - the 1076 device
should be identified as a 82541GI Gigabit Ethernet Controller or a
PRO/1000 MT

>driver: e1000
>desc: "Unknown vendor|Generic e1000 device"
>vendorId: 8086
>deviceId: 1027

and that should be a 82545GM Gigabit Ethernet Controller or a PRO/1000
MF Server Adapter(LX). None the less, the e1000 is the correct driver for
both of these cards, and the e100 is correct for the 82801BD PRO/100 VE
NIC.

Old guy

On 26 mayo, 22:41, ibupro...@painkiller.example.tld (Moe Trin) wrote:
> On Mon, 26 May 2008, in the Usenet newsgroup comp.os.linux.networking, in
> article <52633167-2a60-4a3c-89c4-e80c4c64f...@f36g2000hsa.googlegroups.com>,

> Fionn wrote:

> >recently I have the same error in two different machines running
> >RedHat Linux 9 (kernel 2.4.20-8).

> That's the original kernel on an unmaintained 5 year old system.
> There were at least 9 kernel errata during the supported life, and
> three more backports - ending with 2.4.20-46.9.legacy in March 2006.

For some hardware dependencies (a MOXA multiport communication card),
I must use that O.S. with that kernel.

> >All the system was working properly untill the other day the machine
> >made a reboot (due to the softdog)

> Both systems suffered the same fault at the same time???

Yep, in a period of one week both machines suffered the same fault.
And they were working nice since several weeks.
They have no connection to internet (only local network between my
machines).

> >and the interfaces' configuration became wrong: the system couldn't
> >find the eth2 interface and also it tried to asign eth0 a e1000
> >module, but it couldn't, so it didn't result in a good rising of the
> >interface and failed to be up. Indeed, the e100 module was not loaded
> >(consulting via lsmod)

> That's usually a hardware or BIOS problem, where the e100 NIC isn't
> being found for some reason. *Look at the boot messages in
> /var/log/messages relating to finding the cards. *

The only thing that /var/log/messages told me about network devices is
that the interface eth2 (fiber card) could not be found, and so it
didn't appear when the system was already booted.

>RH9 had a piece of
> crap application called 'kudzu' that was meant to reconfigure the
> system when hardware changed. I usually uninstalled that package as
> my hardware wasn't being changed every time the systems reboot.

I tried to launch kudzu in order to recognize the lost NICs, but it
didn't tell me anything. That's right, since we have seen that the
NICs are right configured in hwconf file.

I also do that. I always switch the kudzu service off from the boot of
the system.
But I realized that, when the error appeared, kudzu was configured (in
the forst machine) to start on level 5 (the one I use), although it
hadn't found anything new and so there was no change in hwconf file.

> >driver: e1000
> >desc: "Unknown vendor|Generic e1000 device"
> >vendorId: 8086
> >deviceId: 1076

> That's another indication that the kernel is obsolete - the 1076 device
> should be identified as a 82541GI Gigabit Ethernet Controller or a
> PRO/1000 MT

> >driver: e1000
> >desc: "Unknown vendor|Generic e1000 device"
> >vendorId: 8086
> >deviceId: 1027

> and that should be a 82545GM Gigabit Ethernet Controller or a PRO/1000
> MF Server Adapter(LX). None the less, the e1000 is the correct driver for
> both of these cards, and the e100 is correct for the 82801BD PRO/100 VE
> NIC.

Exact, that's the fiber card (82545GM).


I have tried to manually change hwconf file, as I've seen that in
other computers was written the complete interface name (eth0, eth1,
eth2).
Now, I have this in hwconf:
-------------------------------------------------
class: NETWORK
bus: PCI
detached: 0
device: eth0
driver: e100
desc: "Intel Corp.|82801BD PRO/100 VE (LOM) Ethernet Controller"
vendorId: 8086
deviceId: 1039
subVendorId: 8086
subDeviceId: 103a
pciType: 1
-
class: NETWORK
bus: PCI
detached: 0
device: eth1
driver: e1000
desc: "Unknown vendor|Generic e1000 device"
vendorId: 8086
deviceId: 1076
subVendorId: 8086
subDeviceId: 1076
pciType: 1
-
class: NETWORK
bus: PCI
detached: 0
device: eth2
driver: e1000
desc: "Unknown vendor|Generic e1000 device"
vendorId: 8086
deviceId: 1027
subVendorId: 8086
subDeviceId: 1027
pciType: 1
--------------------------------------------

When I reboot the computer (not just restart network services), the
interfaces were just there and right configured. I could work with
them.
The next test I will make is deleting the hwconf (as I don't work with
kudzu detecting new hardware) and see if everything works fine without
that file.

Now, I must know if there was a problem with kudzu in that kernel and
a missconfiguration of the hwconf, for making sure that if I also
change that file in all my computers, there will be no more problems.


> * * * * Old guy

On Tue, 27 May 2008, in the Usenet newsgroup comp.os.linux.networking, in
article <2d718185-ced4-45fa-9657-359b49b58a11@m36g2000hse.googlegroups.com>,
Fionn wrote:

NOTE: Posting from groups.google.com (or some web-forums) dramatically
reduces the chance of your post being seen. Find a real news server.

>(Moe Trin) wrote:

>> Fionn wrote:

>>>RedHat Linux 9 (kernel 2.4.20-8).

>> That's the original kernel on an unmaintained 5 year old system.

>For some hardware dependencies (a MOXA multiport communication card),
>I must use that O.S. with that kernel.

Oh, I hate those kinds of problems. kernel.org is still maintaining
the 2.4.x kernel, and the latest version there is 2.4.36.4 released
about 3 weeks ago.

>> Both systems suffered the same fault at the same time???
>
>Yep, in a period of one week both machines suffered the same fault.
>And they were working nice since several weeks.

Stretching credibility for it to be a hardware or BIOS fault, but that
is the "normal" problem. Were there any other changes in hardware?

>They have no connection to internet (only local network between my
>machines).

That's good, as the kernel errata were for at least two security
problems.

>> That's usually a hardware or BIOS problem, where the e100 NIC isn't
>> being found for some reason. <A0>Look at the boot messages in
>> /var/log/messages relating to finding the cards. <A0>
>
>The only thing that /var/log/messages told me about network devices is
>that the interface eth2 (fiber card) could not be found, and so it
>didn't appear when the system was already booted.

Is there any boot messages in the older log files? ('logrotate' is
usually set to rotate /var/log/messages every Sunday ~04:00, and the
default used to be to keep four weeks of such logs.)

>> RH9 had a piece of crap application called 'kudzu' that was meant to
>> reconfigure the system when hardware changed. I usually uninstalled
>> that package as my hardware wasn't being changed every time the
>> systems reboot.

>I also do that. I always switch the kudzu service off from the boot of
>the system.

I was never able to understand the rational for that program.

>I have tried to manually change hwconf file, as I've seen that in
>other computers was written the complete interface name (eth0, eth1,
>eth2).
>Now, I have this in hwconf:

>When I reboot the computer (not just restart network services), the
>interfaces were just there and right configured. I could work with
>them.

OK!

>The next test I will make is deleting the hwconf (as I don't work with
>kudzu detecting new hardware) and see if everything works fine without
>that file.

I'd rename the file, but that's just my paranoia ;-)

>Now, I must know if there was a problem with kudzu in that kernel and
>a missconfiguration of the hwconf, for making sure that if I also
>change that file in all my computers, there will be no more problems.

[compton /net/johnstown/redhat/old]$ ls 9*
9-errata.05.01.04.gz 9-legacy.01.23.07.gz
[compton /net/johnstown/redhat/old]$ zgrep kudzu 9*
[compton /net/johnstown/redhat/old]$ zgrep kudzu rpms.9-i386.gz | cut -c30-
314267 Feb 25 23:20 kudzu-0.99.99-1.i386.rpm
112652 Feb 25 23:20 kudzu-devel-0.99.99-1.i386.rpm
[compton /net/johnstown/redhat/old]$

Near as I can tell, kudzu was never updated from the out-of-box version
(the rpms.9-i386.gz is a directory listing from April 2003 when RH9 was
released - the February date looks to be right after the release of the
last 'phoebe' redhat-8.0.94 beta). As mentioned, the kernel had quite
a number of updates - the oldest one listed on the 9-errata.05.01.04.gz
file was the sixth I have records of, but no dates:

redhat-9 07 Apr 03 shrike 2.4.20-8 -> 2.4.20-13.9 -> 2.4.20-18.9
-> 2.4.20-19.9 -> 2.4.20-24.9 ->2.4.20-27.9
-> 2.4.20-28.9 -> 2.4.20-30.9 -> 2.4.20-31.9
-> 2.4.20-42.9.legacy -> 2.4.20-43.9.legacy
-> 2.4.20-46.9.legacy

kernel-2.4.20-28.9.i386.rpm 24-Dec-2003 14:24

Unless you are running into some bizarre Y2K-like problem, it seems
unlikely that this would be a problem in kudzu or the 2.4.20-8, as if
it were going to happen, you'd think it would have happened before now.
My best guess remains some hardware change, but that's only a guess.

Old guy