Bug 153669

What sorts of devices are these?

device: eth0
driver: tg3
desc: "Broadcom Corporation NetXtreme BCM5704 Gigabit Ethernet"
network.hwaddr: 00:00:1A:19:65:C6
vendorId: 14e4
deviceId: 1648
subVendorId: 1022
subDeviceId: 36c0

device: eth1
driver: tg3
desc: "Broadcom Corporation NetXtreme BCM5704 Gigabit Ethernet"
network.hwaddr: 00:00:1A:19:65:C5
vendorId: 14e4
deviceId: 1648
subVendorId: 1022
subDeviceId: 36c0

Before any interfaces are brought up, if you load the module, is ...:C6 eth1,
and ...:C5 eth0?

The system is a Red Hat Enterprise Linux 4 default installation.

From /etc/modprobe.conf:
alias eth0 tg3
alias eth1 tg3
[...]

Yes, I'm just curious if you turn off hotplug, bring down the network, unload
the module, and reload it, if the interfaces are reversed.

Bill for comment #3: eth0 is ...:C6 and eth1 is ...:C5 further on and they 
don't switch. BTW, tell me what you like to see for "turn off hotplug", maybe
I made something wrong.

But testing this I got another way to reproduce my problem, maybe this helps 
you:

1. /etc/sysconfig/network-scripts/ifcfg-eth0 should contain:
--- snipp ---
DEVICE=eth0
BOOTPROTO=static
BROADCAST=192.168.31.255
HWADDR=00:00:1A:19:65:C6
IPADDR=192.168.24.194
NETMASK=255.255.248.0
NETWORK=192.168.24.0
ONBOOT=no
TYPE=Ethernet
--- snapp ---

2. /etc/sysconfig/network-scripts/ifcfg-eth1 should contain:
--- snipp ---
DEVICE=eth1
BOOTPROTO=static
HWADDR=00:00:1A:19:65:C5
TYPE=Ethernet
ONBOOT=yes
IPADDR=10.208.225.47
NETMASK=255.255.225.0
NETWORK=10.208.225.0
BROADCAST=10.208.225.255
GATEWAY=10.208.225.3
--- snapp ---

3. Do a "service network restart" and review the result using "ip addr show";
   everything is okay, eth1 is set up.
4. Do "rmmod tg3" and after that a "modprobe tg3"; review it again, eth0 and 
   eth1 are set up.
5. Do a further "service network restart" and review again - everything okay,
   as expected.

The interface being brought up on module re-load is due to hotplug. The network
script disables hotplug (echo /bin/true > /proc/sys/kernel/hotplug) while it's
running.

Yes, I saw the difference between enabled and disabled hotplug. But ...:C6 
always was eth0 and ...:C5 was eth1 - they didn't change at any time.

I am having what I believe is the same problem, where ONBOOT=no is ignored on my eth0:x aliases. If I 
have ifcfg-eth0 and ifcfg-eth0:0, where ifcfg-eth0:0 has ONBOOT=no set, eth0:0 is brought up anyway 
at boot. This is with initscripts-7.31.18.EL-1.

That's not the same issue... aliases use ONPARENT=(yes|no).

I opened a new bug and reported it there instead. (bug 155497)

Bill:

I do not have a problem with onboot, but I wanted to report what I see using the
"turn off hotplug & reload module" test you described.

I have two bugs open currently.  Bug 155522 describes problems that appear to be
related most closely to system-config-network-tui.  Bug 156402 describes a weird
ethernet device name, and the results of my test of the "hotplug off & reload" test.

As most directly relates to the present bug, the results of the "hotplug off &
reload" test were that eth0 got the first MAC and eth1 got the second MAC. 
Previously, eth0 got the second MAC and eth1 (misnamed, not "eth1") got the
first MAC.

So maybe the devices are getting misordered when hotplug is involved?  I also
saw misordering on another RHEL4 machine, but I was able to fix it manually
(editing the relevant files).

Created attachment 114210 [details]
patch for this issue

Try this. What I believe is happening is that the fix for token-ring (#90602,
#127984) causes link addresses to be found for more interface types. When the
sit0 interface initializes, it triggers a hotplug event, where it finds a
hwaddr of '0'; since that matches the beginning of the hwaddr for one of your
ethernet devices, that is brought up.

Thank you Bill, this patch works for us. Is it now possible to get an official 
update of initscripts (maybe together with other fixes) or is RHEL4 U1 coming 
very soon?

This is currently scheduled for U2.

Fixed in 7.31.24.EL-1 for RHEL 3.

Fixed in 7.93.14.EL-1 for RHEL-4.

*** Bug 163073 has been marked as a duplicate of this bug. ***

An advisory has been issued which should help the problem
described in this bug report. This report is therefore being
closed with a resolution of ERRATA. For more information
on the solution and/or where to find the updated files,
please follow the link below. You may reopen this bug report
if the solution does not work for you.

http://rhn.redhat.com/errata/RHBA-2005-606.html

An advisory has been issued which should help the problem
described in this bug report. This report is therefore being
closed with a resolution of ERRATA. For more information
on the solution and/or where to find the updated files,
please follow the link below. You may reopen this bug report
if the solution does not work for you.

http://rhn.redhat.com/errata/RHBA-2005-607.html

This type of behaviour still exists when using aliased interfaces:

http://bugs.centos.org/view.php?id=2363

Fix:

--- /etc/sysconfig/network-scripts/ifup-aliases.bug 2007-09-27
11:29:40.000000000 +0200
+++ /etc/sysconfig/network-scripts/ifup-aliases 2007-09-27 11:30:23.000000000 +0200
@@ -321,6 +321,7 @@ for FILE in ifcfg-${parent_device}:* ; d
        is_ignored_file "$FILE" && continue
        ini_env;
        . $FILE;
+ [ "${ONBOOT}" = "no" -o "${ONBOOT}" = "NO" ] && continue
        [ -z "$DEVICE" ] && DEVICE=${FILE##ifcfg-}
        [ "$ONPARENT" != "no" -a "$ONPARENT" != "NO" ] && new_interface;
        unset DEVICE