I understood the very basic concept of how to use /etc/network/interfaces, but all I find online are examples, example after example, which I can copy-paste from. What I miss is an explanation of the syntax, an explanation of the meaning of the commands and which order the commands require. I want to understand, because most of the time copy-paste is not enough, because I'm not working on a fresh machine, so I can't just overwrite existing configurations because it would break a lot of stuff. man interfaces was not very helpful since it is written very complicated.

Example questions I have: what does inet in an iface line mean exactly (I could not even find it in the manpage), what does manual in an iface line mean exactly (many examples use it, but according to manpage it needs an extra config file then, which the examples don't present), when do I use or need them? When not? When I create a bridge, what exactly happens to the interfaces?

  • 4
    The man page says what comes after the interface name is the address family that the interface uses. "inet" is the name for IPv4, inet6 for ipv6. There's also ipx, x25, appletalk..., though ifupdown only deal with inet/inet6/ipx as the man page explains. manual means that ifupdown don't do anything about them, you have to do it yourself manually. – Stéphane Chazelas May 8 '14 at 10:25

Well, let’s separate it into pieces, to make it easier to understand /etc/network/interfaces:

Link layer+interface type options (generally the first of each interface stanza and called address family + method by interfaces(5) manpages):

auto interface – Start the interface(s) at boot. That’s why the lo interface uses this kind of linking configuration.

allow-auto interface – Same as auto

allow-hotplug interface – Start the interface when a "hotplug" event is detected. In the real world, this is used in the same situations as auto but the difference is that it will wait for an event like "being detected by udev hotplug api" or "cable linked". See "Related Stuff(hotplug)" for additional info.

These options are pretty much "layer 2" options, setting up link states on interfaces, and are not related with "layer 3" (routing and addressing). As an example you could have a link aggregation where the bond0 interface needs to be up whatever the link state is, and its members could be up after a link state event:

auto bond0
iface bond0 inet manual
        down ip link set $IFACE down
        post-down rmmod bonding
        pre-up modprobe bonding mode=4 miimon=200
        up ip link set $IFACE up mtu 9000
        up udevadm trigger

allow-hotplug eth0
iface eth0 inet manual
        up ifenslave bond0 $IFACE
        down ifenslave -d bond0 $IFACE 2> /dev/null

allow-hotplug eth1
iface eth1 inet manual
        up ifenslave bond0 $IFACE
        down ifenslave -d bond0 $IFACE 2> /dev/null

So, this way I create a link aggregation and the interfaces will be added to it and removed on cable link states.

Most common interface types:

All options below are a suffix to a defined interface (iface <Interface_family>). Basically the iface eth0 creates a stanza called eth0 on an Ethernet device. iface ppp0 should create a point-to-point interface, and it could have different ways to acquire addresses like inet wvdial that will forward the configuration of this interface to wvdialconf script. The tuple inet/inet6 + option will define the version of the IP protocol that will be used and the way this address will be configured (static, dhcp, scripts...). The online Debian manuals will give you more details about this.

Options on Ethernet interfaces:

inet static – Defines a static IP address.

inet manual – Does not define an IP address for an interface. Generally used by interfaces that are bridge or aggregation members, interfaces that need to operate in promiscuous mode (e.g. port mirroring or network TAPs), or have a VLAN device configured on them. It's a way to keep the interface up without an IP address.

inet dhcp – Acquire IP address through DHCP protocol.

inet6 static – Defines a static IPv6 address.


# Eth0
auto eth0
iface eth0 inet manual
    pre-up modprobe 8021q
    pre-up ifconfig eth0 up
    post-down ifconfig eth0 down

# Vlan Interface
auto vlan10
iface vlan10 inet static
        vlan-raw-device eth0
        ip_rp_filter 0

This example will bring eth0 up, and create a VLAN interface called vlan10 that will process the tag number 10 on an Ethernet frame.

Common options inside an interface stanza(layer 2 and 3):

address – IP address for a static IP configured interface

netmask – Network mask. Can be ommited if you use cidr address. Example:

iface eth1 inet static

gateway – The default gateway of a server. Be careful to use only one of this guy.

vlan-raw-device – On a VLAN interface, defines its "father".

bridge_ports – On a bridge interface, define its members.

down – Use the following command to down the interface instead of ifdown.

post-down – Actions taken right after the interface is down.

pre-up – Actions before the interface is up.

up – Use the following command to up the interface instead of ifup. It is up to your imagination to use any option available on iputils. As an example we could use up ip link set $IFACE up mtu 9000 to enable jumbo frames during the up operation(instead of using the mtu option itself). You can also call any other software like up sleep 5; mii-tool -F 100baseTx-FD $IFACE to force 100Mbps Full duplex 5 seconds after the interface is up.

hwaddress ether 00:00:00:00:00:00 - Change the mac address of the interface instead of using the one that is hardcoded into rom, or generated by algorithms. You can use the keyword random to get a randomized mac address.

dns-nameservers – IP addresses of nameservers. Requires the resolvconf package. It’s a way to concentrate all the information in /etc/network/interfaces instead of using /etc/resolv.conf for DNS-related configurations. Do not edit the resolv.conf configuration file manually as it will be dynamically changed by programs in the system.

dns-search example.net – Append example.net as domain to queries of host, creating the FQDN. Option domain of /etc/resolv.conf

wpa-ssid – Wireless: Set a wireless WPA SSID.

mtu - MTU size. mtu 9000 = Jumbo Frame. Useful if your Linux box is connected with switches that support larger MTU sizes. Can break some protocols(I had bad experiences with snmp and jumbo frames).

wpa-psk – Wireless: Set a hexadecimal encoded PSK for your SSID.

ip_rp_filter 1 - Reverse path filter enabled. Useful in situations where you have 2 routes to a host, and this will force the packet to come back from where it came(same interface, using its routes). Example: You are connected on your lan( and you have a dlna server with one interface on your lan( and other interface on dmz to execute administrative tasks( During a ssh session from your computer to dlna dmz ip, the information needs to come back to you, but will hang forever because your dlna server will try to deliver the response directly through it's lan interface. With rp_filter enabled, it will ensure that the connection will come back from where it came from. More information here.

Some of those options are not optional. Debian will warn you if you put an IP address on an interface without a netmask, for example.

You can find more good examples of network configuration here.

Related Stuff:

Links that have information related to /etc/network/interfaces network configuration file:

  • Thanks, this helps much. Does inet has anything to do with "internet"? I read somehwere that it just means "IPv4" and inet6 means "IPv6", but the term "inet" is really confusing when you see it and confuse it with short for "internet". And the names one uses, like eth0, are these defined or can I chose them myself? If I have more than one LAN adapter, how do I tell which eth belongs to which LAN adapter? – Foo Bar May 9 '14 at 12:48
  • 2
    Both. Together they define method of ip address aquisition(dhcp, static, ppp scripts, no address) and version of ip protocol(inet = v4 and inet6 = v6). The interface names pretty much depends on the distribution you are using, and the way udev is configured. Debian uses eth* and wlan* to cable and wireless interfaces. Fedora uses the biosdevname scheme where em0 is the first built-in ethernet interface on your motherboard, and p<slot>p<eth port> the name for a PCI nic. You can fix network names at /etc/udev/rules.d/70-persistent-net.rules or create alias on interfaces. – user34720 May 9 '14 at 13:05
  • 2
    Regarding "gateway - the default gateway of a server. be carefull to use only one of this guy." Is that one per stanza or one gateway interface among all stanzas? – ctbrown Dec 12 '14 at 14:17
  • 2
    Should be one gateway among all stanzas, or, you will have trouble with your default route. To work with multi gateway / multi link, some thinkering with iptables + package marking + ip rules will be needed. – user34720 Jan 20 '15 at 11:26
  • One small query, if in this file, I am having iface eth0 inet static and then if I write some IP against address field, does this mean that this particular IP that I just wrote would get bonded to my system's MAC address? – k10 Mar 26 '15 at 8:25

I would also add that:

  • interfaces is for ifup/ifdown services.
  • When u use allow-hotplug it will not start with ifup/ifdown, bcos u need to use flag --allow=hotplug.

U may track what is happening with ifup by using --verbose flag.

Im not 100% shure of this, but it seems that basically when ifup is called u will execute all from /etc/network/interfaces with up ... if not stated otherwise.

Im not shure how it relates to : service networking ...

Would be nice if someone would point whats happening after call:

service networking restart

in relation to ifup/ifdown.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.