[CentOS-devel] Ethernet device __tmpXXXXXXXXXX

Thu Jun 26 23:14:46 UTC 2008
Mark Deric <laguna at laguna1.com>

I have a CentOS 5 system that started exhibiting the unusual behavior similar to that described in CentOS bug #2791.

I my case, the interface that used to be eth0, with modprobe.conf line "alias eth0 8139too" and lspci listing "02:09.0 Ethernet controller: Realtek Semiconductor Co., Ltd. RTL-8139/8139C/8139C+ (rev 10)", now comes up like the bug symptom.  After boot, ifconfig shows

__tmp1654060624 Link encap:Ethernet HWaddr 00:1D:09:6A:D8:1C
          BROADCAST MULTICAST MTU:1500 Metric:1

Humorously enough, it is possible to manually configure the interface using that name.  And, using some "set -x" lines in the init scripts, I learned that this does not happen at boot time because the script cannot associate __tmp1654060624 with /etc/sysconfig/network-scripts/ifcfg-eth0.  In the scripts the basic breakdown is that the "ip -o link" command returns the bogus __tmp device name; I'm not sure why, or how to fix this.  A better workaround to getting the interface configured is to execute "ip link set __tmp1654060624 name eth0; service network restart"

Unlike the bug, my setup always boots to a bogus device name (though the 10 digit number following __tmp always changes).

I'm not sure what changed from when it booted to eth0 just fine (though I did have a power outage; but forced the fsck and everything, except this seems fine).  The part I don't get is how the boot time "ip -o link" command gets the bogus device name.

I'd appreciate any help.