vnetbuild.conf - VNetBuild configuration file
host *ID* dev *DEVICE* [ *ID*/*PAIRDEV* ] [ *IP*/*MASK*... ] ... bridgedev *BRIDGE* [ *DEVICE*... ] [ *IP*/*MASK*... ] ... route *ROUTECMD* ... pre_up *DEVICE* *CUSTOMCMD* ... exec *CUSTOMCMD* ... ... switch *ID* dev *DEVICE* [ *ID*/*PAIRDEV* ] ... pre_up *DEVICE* *CUSTOMCMD* ... exec *CUSTOMCMD* ... ...
There is no default configuration file for vnetbuild(1); one must always be specified on the command line.
The configuration file defines a set of namespaces that will be operated on.
VNetBuild defines two types of namespace, a
host and a
switch. Any number of each may be specified, with any number of configuration statements in each.
switchnamespace. VNetBuild provides the distinction to make it easy build full virtual networks.
Namespace definitions come in two types,
switch. Simply provide a simple unique alphanumeric ID. Any subsequent statements apply to this namespace until the next
host definition is designed to work like a physical machine. It allows you to specify any number of
dev entries for network interfaces, with their IP addresses. You can also define any number of Linux bridges with
bridgedev to add your defined interfaces to.
host also allows any number of custom
exec commands for extensibility and provides a
route statement to deal with the common case of wanting to add network routes to the host.
switch definition is designed to work like a physical network switch. It allows you to add any number of
dev entries (and also custom
exec commands for extensibility) but nothing else.
dev entries in a
switch may only specify device names, they cannot have an IP address associated. A
switch has a bridge automatically created in it and all
dev entries are automatically added to it.
Define a virtual ethernet device, DEVICE in a
Devices must exist in pairs. A
dev must first be defined unpaired in a namespace, then some subsequent
dev must define the pair:
host a dev veth0 host b dev vppp0 a/veth0
Any DEVICE name which is acceptable to the Linux kernel may be used. We recommend sticking to e.g.
vppp0 etc. to make it clear that they are virtual and also how you are thinking of the device in terms of your setup. Devices will be created as type
veth, irrespective of what you call them.
Hosts may optionally specify one or more IP/MASK values which will be applied (along with the calculated broadcast address) automatically, e.g.:
host a dev veth0 10.0.0.1/8 192.168.1.2/24 host b dev vppp0 a/veth0 10.0.0.2/8 192.168.1.3/24
dev may not specify an IP address if it is in a
switch. Switches exist just to tie together multiple devices in hosts, just like a physical network switch.
Define an ethernet bridge, BRIDGE in a
host. These are setup automatically using ip(8) and shown with bridge(8).
A bridge can specify network devices from its own namespace to be automatically added, as well as its own IP address(es).
host a dev veth0 dev veth1 otherns/vdev0 bridgedev vbr0 veth0 veth1 10.0.0.3/8
Devices included in a bridge generally do not need their own IP address (although that is permitted).
Bridges cannot have a pair themselves, but any devices added to a bridge need a pair as usual.
Specify an additional network route for a
Most commonly to add a default route from hosts on a "LAN" to the machine that acts as a gateway, e.g.:
route default via 10.0.0.254
The syntax of ROUTECMD is anything that can fit this pattern:
ip route add ROUTECMD
See ip(8) and ip-route(8) for help adding routes. If you want to do anything more complex than simply adding routes, use the
exec configuration statement.
Execute custom commands in a
switch just before bringing up the specified device. All of the
pre_up statements for a device are combined and executed in the namespace.
In addition to any explicitly defined interfaces, switches have an implicit bridge device called
switch which can also be used in
Bridges always start after other devices, so to run a command after all everything has been created but before any interfaces are up, you can make use of
pre_up on the first defined
See below for some common uses for custom
Execute a custom command in a
switch once the rest of the namespace setup is complete.
Once all the namespaces are created, the final step in setting each one up is to have its
exec statements combined and executed.
It is roughly the equivalent to writing your own script and executing it after
vnetbuild start has finished:
sudo ip netns exec myns ./myscript.sh
See below for some common uses for custom
For the most part it doesn't matter whether these commands are used in
exec operations - the only difference is when they will execute, so e.g. if you want a firewall in place before any interfaces come up then start it from the
pre_up of the first device. If you only want the firewall after all devices are up, put it in
host myfirewall ... exec firehol myfirewall.conf start
Forwarding is not enabled by the Linux kernel when a namespace is first created. This can be easily done for any hosts that need to forward traffic:
host mygateway ... exec echo 1 > /proc/sys/net/ipv4/ip_forward
exec operates in the
mygateway namespace so your host is not affected.
Bridges are created without STP being enabled. To enable STP to ensure loops are not created, the following can be done:
host myhost bridgedev vbr0 ... ... pre_up vbr0 echo 2 > /sys/class/net/vbr0/bridge/stp_state switch myswitch ... pre_up switch echo 2 > /sys/class/net/vbr0/bridge/stp_state
You could also use
brctl stp vbr0 on and
brctl stp switch on instead of setting the values directly. To disable multicast snooping you can use exactly the same method e.g.:
switch myswitch ... pre_up switch echo 0 > /sys/class/net/switch/bridge/multicast_snooping
It is possible to run firehol within a namespace to set up custom
Logs from network namespaces are not included in the normal system logs. To enable iptables logging you must start an instance of ulogd(8) in the namespace and use ULOG or NFLOG logging. For FireHOL, that means set
FIREHOL_LOG_MODE=NFLOG. Note that NFLOG only works with ulogd version 2.
The default configuration for ulogd(8) is
/etc/ulogd.conf. Assuming the default place it will write iptables logs to is
/var/log/ulog/syslogemu.log (otherwise change the
sed command as required), it is simple to set up per-namespace logging:
host mygateway ... exec sed 's:/var/log/ulog/syslogemu.log:/var/log/ulog/mygateway.log:' /etc/ulogd.conf > $NSTMP/ulogd.conf exec /usr/sbin/ulogd -d -c $NSTMP/ulogd.conf
-d flag to ulogd(8) makes it become a daemon; when
vnetbuild stop executes it will automatically kill any programs running in the namespaces is is stopping, which includes the logging daemon.
The configuration file will get cleaned as soon as
vnetbuild start is finished. To be able to access such files you need to write them to a location not under
$NSTMP or create them outside the
vnetbuild configuration altogether.
A simple LAN arrangement with two hosts, one of which is a gateway to third host:
host host01 dev veth0 10.0.0.1/8 dev vppp0 192.168.0.1/24 exec echo 1 > /proc/sys/net/ipv4/ip_forward route default via 192.168.0.1 host host02 dev veth0 10.0.0.2/8 route default via 10.0.0.1 switch lan dev d01 host01/veth0 dev d02 host02/veth0 host extern01 dev veth0 host01/vppp0 192.168.0.254/24 route default via 192.168.0.1 exec echo 1 > /proc/sys/net/ipv4/ip_forward
When created, the namespaces setup by
vnetbuild are completely disconnected from any real network. There is no way of defining such a connection in the
vnetbuild configuration as allowing it would lead to conflicts with the normal network setup tools and configuration files in most distributions.
It is possible to arrange your network so you can connect real devices into one or more network namespaces. For the general approach see this mailing list post.