Using ifplugd on a Server Host
By default, Linux (other than Cumulus Linux, as this article discusses server hosts) does not remove the IP address on connected interfaces. Further, static routes do not associate the interface when a link goes down (as in NO-CARRIER). Read more about determining the administrative/physical state on Linux.
To understand what this means, first review this persistent configuration of an Ubuntu 14.10 VM, which gets stored in /etc/network/interfaces (not network-manager):
auto eth1 iface eth1
inet static
address 5.5.5.5
netmask 255.255.255.0
broadcast 5.5.5.255
up route add -net 3.3.0.0/16 gw 5.5.5.1 dev eth1
On this VM, eth1 is UP because the output below indicates LOWER_UP. In addition, its state is UP.
user@ubuntu:~$ ip link show eth1
3: eth1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP mode DEFAULT group default qlen 1000 link/ether 00:0c:29:8e:9e:3a brd ff:ff:ff:ff:ff:ff
In looking at the routing table:
user@ubuntu:~$ ip route show
default via 192.168.72.2 dev eth0
3.3.0.0/16 via 5.5.5.1 dev eth1
5.5.5.0/24 dev eth1 proto kernel scope link src 5.5.5.5
169.254.0.0/16 dev eth1 scope link metric 1000
192.168.72.0/24 dev eth0 proto kernel scope link src 192.168.72.136
Notice the following:
- The connected route of 5.5.5.0/24 on eth1 (because the assigned address is 5.5.5.5 and is on the 255.255.255.0 subnet)
- A 3.3.0.0/16 static route configured on this Ubuntu VM
Next, simulate no connection on eth1 (for example, someone pulled out the cable). You can do this by assigning the Ethernet device in the VM to a dead interface:
user@ubuntu:~$ ip link show eth1
3: eth1: <NO-CARRIER,BROADCAST,MULTICAST,UP> mtu 1500 qdisc pfifo_fast state DOWN mode DEFAULT group default qlen 1000
link/ether 00:0c:29:8e:9e:3a brd ff:ff:ff:ff:ff:f
The NO-CARRIER and the DOWN state indicate that this link is administratively up, but physically down. Now check the routing table:
user@ubuntu:~$ ip route show
default via 192.168.72.2 dev eth0
3.3.0.0/16 via 5.5.5.1 dev eth1
5.5.5.0/24 dev eth1 proto kernel scope link src 5.5.5.5
169.254.0.0/16 dev eth1 scope link metric 1000
192.168.72.0/24 dev eth0 proto kernel scope link src 192.168.72.136
Notice that nothing changed in the configuration. While this is helpful in some situations — for example, your Web connections do not reset when your wireless card moves between APs and momentarily goes down — it is not good in others and can cause traffic to be black-holed. For servers this is not usually a problem unless you are doing something fancy like having dual gateways (two or more equal cost routes). In this case, the server attempts to load balance on the dead link as the route was not removed for that interface, resulting in 50% of your connections failing as they try to use the link that is dead.
Demoing ifplugd
With the link on the Ubuntu VM reset and the link set to UP in VMware Fusion (the hypervisor), you can start ifplugd.
user@ubuntu:~$ sudo service ifplugd start
* Network Interface Plugging Daemon...
* start eth1...
[ OK ]
user@ubuntu:~$
user@ubuntu:~$ service ifplugd status
* eth1: ifplugd process for device eth1 running as pid 4512.
user@ubuntu:~$ ip link show eth1 3: eth1: <NO-CARRIER,BROADCAST,MULTICAST,UP> mtu 1500 qdisc pfifo_fast state DOWN mode DEFAULT group default qlen 1000 link/ether 00:0c:29:8e:9e:3a brd ff:ff:ff:ff:ff:ff
At this point, the cable has been “cut,” which you can see in the routing table:
user@ubuntu:~$ ip link show eth1
3: eth1: <NO-CARRIER,BROADCAST,MULTICAST,UP> mtu 1500 qdisc pfifo_fast state DOWN mode DEFAULT group default qlen 1000
link/ether 00:0c:29:8e:9e:3a brd ff:ff:ff:ff:ff:ff
user@ubuntu:~$ ip route show
default via 192.168.72.2 dev eth0
169.254.0.0/16 dev eth0 scope link metric 1000
192.168.72.0/24 dev eth0 proto kernel scope link src 192.168.72.136
Notice both routes no longer appear in the output above. Next, the cable is back in, so the link goes back to administratively UP and physically UP:
user@ubuntu:~$ ip link show eth1
3: eth1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP mode DEFAULT group default qlen 1000
link/ether 00:0c:29:8e:9e:3a brd ff:ff:ff:ff:ff:ff
user@ubuntu:~$ ip route show
default via 192.168.72.2 dev eth0
3.3.0.0/16 via 5.5.5.1 dev eth1
5.5.5.0/24 dev eth1 proto kernel scope link src 5.5.5.5
169.254.0.0/16 dev eth0 scope link metric 1000
192.168.72.0/24 dev eth0 proto kernel scope link src 192.168.72.136
Notice that everything looks as it did before. ifplugd has helped the server act more like a router (or layer 3 switch) and enables the connected and static routes to disappear.
Configuring ifplugd on Ubuntu 14.10
The following examples use ifupdown (not ifupdown2 in NVIDIA and not network-manager.
Run
apt-get updateto grab the latest package information:user@ubuntu:~$ sudo apt-get updateInstall
ifplugd:user@ubuntu:~$ sudo apt-get install ifplugd Reading package lists... Done Building dependency tree Reading state information... Done Suggested packages: waproamd The following NEW packages will be installed: ifplugd 0 upgraded, 1 newly installed, 0 to remove and 0 not upgraded. Need to get 0 B/65.1 kB of archives. After this operation, 276 kB of additional disk space will be used. Preconfiguring packages ... Selecting previously unselected package ifplugd. (Reading database ... 193504 files and directories currently installed.) Preparing to unpack .../ifplugd_0.28-19ubuntu1_amd64.deb ... Unpacking ifplugd (0.28-19ubuntu1) ... Processing triggers for man-db (2.7.0.2-2) ... Processing triggers for ureadahead (0.100.0-16) ... user@ubuntu:~$Modify the default configuration for
ifplugd. Usevi,nano, or your preferred text editor. You should see the following output if you are runningifplugdonly for eth1 (normally you separate two or more interfaces with spaces, for example,eth1 eth2 eth3):cat /etc/default/ifplugd #(REMOVE GIANT COMMENT BLOCK FOR BREVITY) INTERFACES="" HOTPLUG_INTERFACES="eth1" ARGS="-q -f -u0 -d0 -w -I" SUSPEND_ACTION="none"Explanation of flags:
Flag Description -q Do not call the script for network shutdown when the daemon quits. -f Ignore detection failures, retry instead. This option treats failures as “no link.” -u0 Specifies the delay for configuring the interface. -d0 Specifies the delay for de-configuring the interface. -w When daemonizing (creating a background process), wait until the background process finishes with the initial link beat detection. When you enable this option, the parent process returns the link status on exit. 2 means it detects a link beat, 3 means it does not detect a link beat; everything else is an error. -I Do not exit on a nonzero return value of the program executed on link change. -M Do not fail when the network interface is not available. Instead, use netlinkto monitor device availability. The is useful for PCMCIA devices and similar.To start
ifplugd, use theservice ifplugd startcommand:user@ubuntu:~$ sudo service ifplugd start * Network Interface Plugging Daemon... * start eth1... [ OK ]To check if
ifplugdis running, use theservice ifplugd statuscommand:user@ubuntu:~$ sudo service ifplugd status * eth1: ifplugd process for device eth1 running as pid 16327. * eth1: ifplugd process for device eth1 running as pid 16327.