lcfg-network

Package release details for current stable version under FC5 FC6

Package Changelog

Component documentation

Component man page for current stable version under FC5 FC6

Known Issues

• The network.offline resource, as documented in the man page, is not implemented.
• The component will not remove configuration files for interfaces which have dropped from the network.interfaces list.

---

Multi homed machines

Alias (or clone) interfaces

It is possible to configure a network interface to have two IP addresses on the same subnet using alias (or clone) interfaces. The following code adds an alias interface eth0:1

!network.interfaces          mADD(eth01)
network.device_eth01      eth0:1
network.ipaddr_eth01      129.215.212.48
network.hostname_eth01 stoater-etaps

Note that you must specify either an IP address,or a DNS name which resolves to a single IP address, for the ipaddr tags of all interfaces which are on the same subnet. This means that in this example, the following is also required for the primary interface.

!network.ipaddr_eth0  mSET(129.215.212.47)

Channel bonding (eg for HA)

Assuming that you have two physical interfaces eth0 and eth1 and that both are configured such that they are visible (eg with netstat -i),

Create the bond0 interface :-

!hardware.modlist       mADD(bond0 bond0opts)
hardware.mod_bond0      alias bond0 bonding
hardware.mod_bond0opts  options bond0 mode=1 miimon=100 primary=eth0

This will create an active-backup bond interface, with eth0 being the preferred link. Remove the primary=eth0 setting if you don't care which interface is used (eg when they're both the same speed).

Bond eth0 and eth1 under bond0 :-

!network.interfaces     mADD(bond0 eth1)
network.hostname_bond0  <%profile.node%>
network.master_eth0    bond0
network.master_eth1    bond0

Note that you may need to set the network.netmask_bond0 resource to an appropriate value (the default is 255.255.255.0).

You can monitor the status of the bonded interface by reading /proc/net/bonding/bond0

To simplify things, you can use the <lcfg/options/etherbond.h> header. The following will create a bonded pair from eth0 and eth2 :-

#define LCFG_OPTIONS_ETHERBOND_FIRST eth0  
#define LCFG_OPTIONS_ETHERBOND_SECOND eth2 
#include <lcfg/options/etherbond.h>

VLANs

VLANs are added using the network.interfaces resource (as are physical interfaces). The network.vlan_*tag* resource should be set to yes to indicate that the interface is a VLAN, and the network.physdev_*tag* should be set to the hosting physical interface.

The following adds a VLAN with tag 252 to the eth0 interface and sets the IP address for the VLAN interface to be that for the DNS host bakerloo-e.

/* Add VLAN wire DICE interface on top of eth0 */
!network.interfaces                mADD(vlan252)
network.physdev_vlan252      eth0
network.vlan_vlan252             yes
network.hostname_vlan252     bakerloo-e

Note that reordering is enabled by default under SL6, and disabled by default under SL5.

Configuring to use DHCP

The network component can be instructed to use DHCP for a network interface by setting the ipaddr tag of the interface to DHCP, eg

!network.ipaddr_eth0 mSET(DHCP)

Note that the hostname of the machine is fixed at install time.

Consistent device naming.

From RHEL 7.2 onwards, we have adopted the modern consistent device naming scheme.

On most desktops, the onboard network port is known as em1. However, older desktops may name this port something else. ConsistentNamingSchemeByModel lists the name for each model type.