Bond not keeping MTU of 9000 after ifdown/ifup - See also BONDING_MASTER_UP_ENSLAVE option

  • 7014947
  • 22-Apr-2014
  • 22-Apr-2014

Environment

SUSE Linux Enterprise Server 11 Service Pack 3 (SLES 11 SP3)
SUSE Linux Enterprise Server 11 Service Pack 2 (SLES 11 SP2)

Situation

On SLES 11 SP3 with a bonded interface configured to have the MTU set to 9000,
doing an ifdown/ifup on that bonded interface will result in the MTU setting being reset back
to the default of 1500, unless the BONDING_MASTER_UP_ENSLAVE option is set to yes.

The ifcfg-bonding man page does not mention this symptom/issue/requirement.
ifcfg-bonding at the time of this writing said this: BONDING_MASTER_UP_ENSLAVE This option allows to revert the change and set the master "link up" before enslave. Older kernels required bonding master in "link up" state while enslave. Recent kernels have been adopted to allow enslave in down state and this seems to work better, faster and avoids several (false) problems, because in "link down" state, several async actions aren't started at "set up" or enslave time of the first slave and need to be aborted or repeated later.

Resolution

In the configuration file for the bond (E.g. /etc/sysconfig/network/ifcfg-bond0)
set the BONDING_MASTER_UP_ENSLAVE='yes'

Example Configuration: Network cards: ifcfg-eth0 BOOTPROTO='none' BROADCAST='' ETHTOOL_OPTIONS='' IPADDR='' MTU='' NETMASK='' NETWORK='' REMOTE_IPADDR='' STARTMODE='auto' USERCONTROL='no' ifcfg-eth1 BOOTPROTO='none' BROADCAST='' ETHTOOL_OPTIONS='' IPADDR='' MTU='' NETMASK='' NETWORK='' REMOTE_IPADDR='' STARTMODE='auto' USERCONTROL='no' ifcfg-bond0 BONDING_MASTER='yes' BONDING_MODULE_OPTS='mode=802.3ad miimon=100' xmit_hash_policy='layer3+4' BONDING_SLAVE0='eth0' BONDING_SLAVE1='eth1' BOOTPROTO='none' BROADCAST='' ETHTOOL_OPTIONS='' IPADDR='' MTU='9000' NAME='' NETWORK='' REMOTE_IPADDR='' STARTMODE='auto' USERCONTROL='no' PREFIXLEN='16' BONDING_MASTER_UP_ENSLAVE='yes' Restart the network or ifdown/ifup the bonded interface to have the new setting take effect.

Feedback service temporarily unavailable. For content questions or problems, please contact Support.