iproute2 is the Linux networking toolkit that replaced legacy tools (ifconfig, vconfig, brctl, route, arp etc.). Those tools are only kept for compatibility with old scripts and do not provide access to a lot of newer networking features of the Linux kernel.
It originally written by Alex Kuznetsov and is now maintained by Stephen Hemminger.
Most of the networking functionality is unified in the ip command. There’s also tc for managing traffic policies (QoS), bridge for managing software bridge interfaces, and ss (a netstat replacement).
Those commands are usually shipped in a package called iproute2 or iproute. Most Linux distributions install it by default these days. The ip command is sometimes installed in /sbin and thus may not be in the $PATH of unprivileged users by default.
Historically, documentation has been a weak side of iproute2. The official man pages list available options but don’t give almost any usage examples. That need has been addressed by third-party documentation.
This document aims to provide a comprehensive but easy to use guide to the ip and bridge commands, and some information about ss. Documenting tc in this style would be a separate big project.
The document is task-centered: it tells you how to do different tasks using iproute2 commands instead of listing available subcommands.
This document is maintained by Daniil Baturin and distributed under CC-BY-SA 4.0 — a strong copyleft, free culture license.
Contributions are always welcome; you can find the source files at github.com/dmbaturin/iproute2-cheatsheet.
You can also show your support by buying the maintainer a metaphorical coffee.
This document is provided “as is”, without any warranty. The authors are not liable for any damage related to using it. As usual, think before you type, and think twice before hitting the Enter key.
Metasyntactic variables are written in a shell-like syntax, ${something}. Optional command parts are in square brackets. Mandatory arguments are in angle brackets.
All commands that change any settings require root privileges. Commands that just display information generally do not require special privileges.
There are configuration files in /etc/iproute2, mainly for assigning symbolic names to network stack entities such as routing tables. Those files are re-read every time you run the ip command, so you don’t need to do anything to apply the changes.
Any ip command can be abbreviated. For example, ip address add 192.0.2.1/24 dev eth0 can be written ip addr a 192.0.2.1/24 dev eth0 or even ip a a 192.0.2.1/24 dev eth0.
In some cases, you can even omit words. For example, show and list words are always fine to omit: ip address is equivalent to ip address show and ip address list.
Note that the abbreviation system is not always consistent. The dev keyword in ip a a 192.0.2.1/24 dev eth0 cannot be abbreviated, even though every other word can be.
This document intentionally gives all commands in their fullest form for better readability. It’s also a good idea to use full forms in scripts because readers may not be familiar with abbreviations, and code is read much more often than it’s written.
A common complaint about distributions removing ifconfig is that it forces people to rewrite scripts. However, iproute2 is better for scripting since it supports machine-readable output.
It provides the following output options:
Here is a comparison of outputs (–json –pretty and –json –brief are omitted to save space):
iproute2 accepts both dotted decimal masks and prefix length values. That is, both 192.0.2.10/24 and 192.0.2.10/255.255.255.0 are acceptable formats.
All show commands can be used with -4 or -6 options to show only IPv4 or IPv6 addresses.
Show only statically configured addresses:
Show only addresses learnt via autoconfiguration:
You can add as many addresses as you want.
If you add more than one address, your machine will accept packets for all of them. The first address you add becomes a “primary address”. The primary address of an interface it’s used as the source address for outgoing packets by default. All additional addresses you set will become secondary addresses.
The label must start with the interface name followed by a colon due to some backward compatibility issues, otherwise you’ll get an error. Keep the label shorter than sixteen characters, or else you’ll get this error: RTNETLINK answers: Numerical result out of range.
For IPv6 addresses, this command has no effect. It will add the address correctly but will ignore the label.
An interface name is required—the kernel will not try to automatically guess which interface you want to remove that address from. Such a guess would not always be unambiguous: Linux does allow the same address to be configured on multiple interfaces, and it has valid use cases (in the Cisco world, this is known as “unnumbered interfaces”).
By default, this command removes both IPv4 and IPv6 addresses. If you want to remove only IPv4 or IPv6 addresses, use ip -4 address flush or ip -6 address flush.
There is no way to swap primary and secondary IPv4 addresses or explicitly set a new primary IPv4 address. Try to always set the primary address first.
If the sysctl variable net.ipv4.conf.${interface}.promote_secondaries is set to 1, when you delete a primary address, the first secondary address becomes primary. You can enable this behaviour globally with net.ipv4.conf.default.promote_secondaries=1.
Note that when promote_secondaries is set to 0, removing a primary address will also remove all secondary addresses from its interface. This setting varies between Linux distributions, so be careful to check it before attempting to change a primary address.
Secondary IPv6 addresses are always promoted to primary if a primary address is deleted.
This command supports both American (ip neighbor) and British (ip neighbour) spelling variants.
All “show” commands support -4 and -6 options to view only IPv4 (ARP) or IPv6 (NDP) neighbors. By default, all neighbors are displayed.
Examples: ip neighbor show dev eth0
Examples: ip neighbor flush dev eth1
Examples: ip neighbor add 192.0.2.1 lladdr 22:ce:e0:99:63:6f dev eth0
One use case for it is a form of data link layer security. You can disable ARP on an interface completely and add MAC addresses of authorized devices by hand.
Examples: ip neighbor delete 192.0.2.1 lladdr 22:ce:e0:99:63:6f dev eth0
Allows you to delete a static entry or get rid of an automatically learnt entry without flushing the table.
“Link” is another term for a network interface. Commands from the ip link family perform operations that are common for all interface types, like viewing link information or changing the MTU.
Historically, the ip link command could not create tunnels (IPIP, GRE etc.), VXLAN links, or L2TPv3 pseudowires. Starting from at least iproute2 3.16, it could create anything except L2TPv3 interfaces, and this remains true as of iproute2 5.7. A lot of time, old commands for specific interface types are still more convenient to use, though.
Note that the Linux kernel allows arbitrary, even non-ASCII names for network interfaces. It’s better to stick with alphanumeric because userspace programs (like iptables) may not be so forgiving.
These commands are equivalent.
Note: virtual links (tunnels, VLANs, etc.) are always created in the “down” state. You need to bring them up to start using them.
Examples: ip link set dev eth0 alias "LAN interface".
Link aliases show up in the ip link show output, like this:
The kernel limits the length of link names to 15 characters. In order to overcome this limitation, alternative names can be added to a link.
Examples: ip link property add dev eth0 altname eno1 altname enp3s0
Alternative names can be used to refer to the link in iproute commands.
Examples: ip link set dev eth0 name lan
Note that you can’t rename an active interface. You need to bring it down before renaming it.
A link-layer address is a pretty broad concept. The most known example is the MAC address of an Ethernet device. To change a MAC address, you would need something like ip link set dev eth0 address 22:ce:e0:99:63:6f.
Examples: ip link set dev tun0 mtu 1480
MTU stands for “Maximum Transmission Unit”, the maximum size of a frame an interface can transmit at once.
Apart from reducing fragmentation in tunnels, this is also used to increase the performance of gigabit ethernet links that support so-called “jumbo frames” (frames up to 9000 bytes large). If all your equipment supports gigabit ethernet, you may want to do something like ip link set dev eth0 mtu 9000.
Note that you may need to configure it on your L2 switches too, some of them have jumbo frames disabled by default.
Obviously, only virtual links can be deleted, like VLANs, bridges, or tunnels. For physical interfaces, this command has no effect.
Unless you really know what you are doing, better don’t touch this option.
One may want to disable ARP to enforce a security policy and allow only specific MACs to communicate with the interface. In this case, neighbor table entries for whitelisted MACs should be created manually, or nothing will be able to communicate with that interface.
In most cases, it’s better to configure MAC policy on an access layer switch, though. Do not change this flag unless you are sure what you are going to do and why.
Examples: ip link add name eth0.110 link eth0 type vlan id 110
The only type of VLAN supported by Linux is IEEE 802.1q VLAN; legacy implementations like ISL are not supported.
You can use any name for a VLAN interface. eth0.110 is a traditional format, but it’s not required.
Any Ethernet-like device can be a parent for a VLAN interface: bridge, bonding, L2 tunnels (GRETAP, L2TPv3…).
VLAN stacking (aka 802.1ad QinQ) is a way to transmit VLAN tagged traffic over another VLAN. The common use case for it is like this: suppose you are a service provider and you have a customer who wants to use your network infrastructure to connect parts of their network to each other. They use multiple VLANs in their network, so an ordinary rented VLAN is not an option. With QinQ you can add a second tag to the customer traffic when it enters your network and remove that tag when it exits, so there are no conflicts, and you don’t need to waste VLAN numbers.
The service tag is a VLAN tag the provider uses to carry client traffic through their network. The client tag is a tag set by the customer.
Note that link MTU for the client VLAN interface is not adjusted automatically; you need to take care of it yourself and either decrease the client interface MTU by at least 4 bytes or increase the parent MTU accordingly.
Standards-compliant QinQ is available since Linux 3.10.
Examples: ip link add name peth0 link eth0 type macvlan
MACVLAN interfaces are like “secondary MAC addresses” on a single interface. They look like normal Ethernet interfaces from the user’s point of view, and handle all traffic for their MAC address.
This can be used for services and applications that can’t handle secondary IP addresses well.
Examples: ip link add name dummy0 type dummy
In Linux, for some strange historical reasons, there’s only one loopback interface. Dummy interfaces are like loopbacks, but there can be many of them.
They can be used for communication inside a single host. A loopback or a dummy interface is also a good place to assign a management address on a router with multiple physical interfaces.
Examples: ip link add name bond1 type bond
Note: This is not enough to configure bonding (link aggregation) in any meaningful way. You need to set up bonding parameters according to your situation. This is far beyond the cheat sheet scope, so consult the documentation.
Bonding interface members are added and removed exactly like bridge ports, with master and nomaster commands.
Example: ip link add ifb10 type ifb
Intermediate functional block devices are used for traffic redirection and mirroring in conjunction with tc. This is also far beyond the scope of this document; consult the tc documentation.
Virtual ethernet (veth) devices always come in pairs and work as a bidirectional pipe: whatever comes into one of them comes out of the other. They are used in conjunction with system partitioning features such as network namespaces and containers (OpenVZ or LXC) for connecting one partition to another.
Examples: ip link add name veth-host type veth peer name veth-guest
Note: depending on the kernel version, those devices may be created in either down or up state. In recent versions (tested in 5.10) they are down. It’s better to assume that they are down and always use ip link set ${intf} up on them.
Link groups are similar to port ranges found in managed switches. You can add network interfaces to a numbered group and perform operations on all the interfaces from that group at once.
Links not assigned to any group belong to group 0 (“default”).
This can be done by assigning it to the default group.
Examples: ip link set dev tun10 group 0
Group names are configured in the /etc/iproute2/group file. The symbolic name “default” for group 0 isn’t an iproute2 built-in; it comes from the default group config file. You can add your own, one per line, following the same ${number} ${name} format. You can have up to 255 named groups.
Once you configured a group name, its number and name can be used interchangeably in ip commands.
After that, you can use that name in all operations, like this:
Use a usual information viewing command with a group ${group} modifier.
Bridge interfaces are virtual Ethernet switches.
You can use them to turn a Linux box into a slow L2 switch, or to enable communication between virtual machines on a hypervisor host.
Note that turning a Linux box into a physical switch isn’t a completely absurd idea, since unlike dumb hardware switches, it can work as a transparent firewall.
You can assign an IP address to a bridge and it will be visible from all bridge ports.
If bridge creation fails, check if the bridge module is loaded.
Bridges in Linux are managed with two different utilities. For creating bridges and adding or removing ports, you will need commands from the ip link family. Many other tasks, such as assigning port VLANs and viewing bridge forwarding tables, require a command called bridge. That command should not be confused with the older brctl utility.
Examples: ip link add name br0 type bridge
Examples: ip link set dev eth0 master br0
An interface you add to a bridge becomes a virtual switch port. It operates only on the data link layer and ceases all network layer operation.
Examples: ip link set dev eth0 nomaster
In many settings, client hosts only need to communicate to the router rather than to other clients. Linux bridges allow marking ports as “isolated” — an isolated port cannot communicate with a non-isolated port.
You can mark a port isolated using the following command:
In this example, eth0 will not be able to communicate directly with eth1 anymore, but will still communicate with the Linux bridge host itself, and with any ports not marked as isolated.
By default, Linux bridges act like unmanaged switches. However, it’s possible to make them VLAN-aware and then set ports as trunk or access ports.
To enable VLANs, add vlan_filtering 1 to the command options. You can also set the VLAN protocol (802.1q or 802.1ad for nested QinQ VLANs) and set the default VLAN value with vlan_default_pvid.
Examples: ip link add name br0 type bridge vlan_filtering 1
Example: bridge vlan add vid 100 dev eth0 pvid untagged
If you don’t specify pvid and/or untagged option, the bridge will assume that the client device assigns VLAN tags itself.
Note: it’s possible to specify a range of VLANs rather than a single tag, like vid 100-200.
Example: bridge vlan add vid 100-200 dev eth0.
Example: bridge -detail vlan show
Note: bridge vlan show command displays all individual VLANs from all ranges by default. If you do something like bridge vlan add vid 500-550 dev eth0, you wil get fifty lines of individual VLANs. To display contiguous ranges like 500-550, add -com/-compressvlans. If you use -detail, VLAN range compression is enabled automatically.
Running ip neighbor show dev br0 will only show MAC addresses of devices that have already communicated with the bridge host. MAC addresses of devices that only communicate with other bridge ports can be viewed using bridge fdb show commands instead.
Example: bridge fdb show dev br0 — show the MAC address table for bridge br0.
TUN and TAP devices allow userspace programs to emulate a network device. The difference between the two is that TAP devices work with Ethernet frames (L2 device), while TUN works with IP packets (L3 device).
There are two types of TUN/TAP devices: persistent and transient. Transient TUN/TAP devices are created by userspace programs when they open a special device and are destroyed automatically when the associated file descriptor is closed. Persistent devices are created with ip commands documented below.
Note: this command can be abbreviated ip tuntap.
This command is the only way to find out if some device is in the TUN or TAP mode.
Add meta information to each packet received over the file descriptor. Very few programs expect this information, and programs that don’t expect it will not function correctly with a device in this mode.
Example: ip tuntap add dev tun1 mode tun pi
Normally packets sent to a TUN/TAP device travel in the same way as packets sent to any other device: they are put in a queue handled by the traffic control engine (which is configured by the tc command). This can be bypassed, thus disabling the traffic control engine for this TUN/TAP device.
Example: ip tuntap add dev tun1 mode tun one_queue
Note: you must specify the mode. The mode is not displayed in ip link show, so if you don’t know if it’s a TUN or a TAP device, consult the output of ip tuntap show.
Tunnels are “network wormholes” that emulate a direct connection over a routed network by encapsulating entire packets into another protocol.
Linux currently supports IPIP (IPv4 in IPv4), SIT (IPv6 in IPv4), IP6IP6 (IPv6 in IPv6), IPIP6 (IPv4 in IPv6), GRE (virtually anything in anything), and VTI (IPv4 in IPsec).
Note that tunnels are created in the DOWN state; you need to bring them up.
In this section ${local endpoint address} and ${remote endpoint address} refer to addresses assigned to physical interfaces, while ${address} refers to an address assigned to a tunnel interface.
This type of tunnel is commonly used to provide an IPv4-connected network with IPv6 connectivity. There are so-called “tunnel brokers” that provide it to everyone interested, e.g., Hurricane Electric’s tunnelbroker.net.
Examples: ip -6 tunnel add tun8 mode ipip6 local 2001:db8:1::1 remote 2001:db8:1::2
This type of tunnel will be widely used only when transit operators phase IPv4 out (i.e., not any time soon).
Just like IPIP6 these ones aren’t going to be widely used any time soon.
Static GRE tunnels are traditionally used for encapsulating IPv4 or IPv6 packets, but the RFC does not limit GRE payloads to L3 protocol packets. It’s possible to encapsulate anything, including Ethernet frames.
However, in Linux, the gre encapsulation refers specifically to L3 devices, while for an L2 device capable of transmitting Ethernet frames, you need to use the gretap encapsulation.
These tunnels can be bridged with other physical and virtual interfaces.
GRE can encapsulate both IPv4 and IPv6 at the same time. However, by default, it uses IPv4 for transport, for GRE over IPv6 there is a separate tunnel mode, ip6gre.
RFC2890 defines “keyed” GRE tunnels. A “key” in this case has nothing to do with encryption; it’s simply an identifier that allows routers to tell one tunnel from another, so you can create multiple tunnels between the same endpoints.
You can also specify keys in a dotted-decimal IPv4-like format.
Note the absence of ${remote endpoint address}. This is the same as “mode gre multipoint” in Cisco IOS.
In the absence of a remote endpoint address, the key is the only way to identify the tunnel traffic, so ${key value} is required.
This type of tunnel allows you to communicate with multiple endpoints by using the same tunnel interface. It’s commonly used in complex VPN setups with multiple endpoints communicating to one another (in Cisco terminology, “dynamic multipoint VPN”).
Since there is no explicit remote endpoint address, it is obviously not enough to just create a tunnel. Your system needs to know where the other endpoints are. In real life, NHRP (Next Hop Resolution Protocol) is used for it.
For testing, you can add peers manually (given remote endpoint uses 203.0.113.6 address on its physical interface and 10.0.0.2 on the tunnel):
You will have to do it on the remote endpoint as well:
Recent kernel and iproute2 versions support GRE over IPv6. Point-to-point with no key:
It should support all options and features supported by the IPv4 GRE described above.
Note that in older iproute2 versions, this command did not support the full delete syntax, only del. Recent versions allow both full and abbreviated forms (tested in iproute2-ss131122).
Note: Apparently, you can’t add a key to a previously unkeyed tunnel. Not sure if it’s a bug or a feature. Also, you can’t change tunnel mode on the fly, for obvious reasons.
L2TPv3 is a tunneling protocol commonly used for L2 pseudowires.
In many distros, L2TPv3 is compiled as a module and may not be loaded by default. If running any ip l2tp command produces errors like RTNETLINK answers: No such file or directory and Error talking to the kernel, you need to load l2tp_netlink and l2tp_eth kernel modules. If you want to use L2TPv3 over IP rather than over UDP, also load l2tp_ip.
Compared to other tunneling protocol implementations in Linux, L2TPv3 terminology is somewhat backwards. You create a tunnel and then bind sessions to it. You can bind multiple sessions with different identifiers to the same tunnel. Virtual network interfaces (by default named l2tpethX) are associated with sessions rather than tunnels.
You can only create static (unmanaged) L2TPv3 tunnels with iproute2. If you want to use L2TP for remote access VPN or otherwise need dynamically created pseudowires, you need a userspace daemon to handle it. That is outside of this document’s scope.
Note: Tunnel identifiers and other settings on both endpoints must match.
Encapsulating L2TPv3 frames in IP rather than in UDP creates less overhead, but may cause problems for endpoints behind a NAT.
Notes: tunnel_id value must match the value of an existing tunnel (iproute2 will not create a tunnel if it doesn’t exist). Session identifiers on both endpoints must match.
Once you create a tunnel and a session, an l2tpethX interface will appear in a DOWN state. Change the state to UP and bridge it with another interface or assign an address to it.
Examples: ip l2tp del session tunnel_id 1 session_id 1
Examples: ip l2tp del tunnel tunnel_id 1
Note: You need to delete all sessions associated with a tunnel before deleting the tunnel itself.
Examples: ip l2tp show tunnel tunnel_id 12
Examples: ip l2tp show session session_id 1 tunnel_id 12
VXLAN is a tunneling protocol designed for distributed switched networks. It’s often used in virtualization setups to decouple the virtual network topology from that of the underlying physical network.
VXLAN can work in either multicast or unicast mode and supports isolating virtual networks using a VNI (virtual network identifier), similar to VLANs in Ethernet networks.
The downside of the multicast mode is that you will need to use a multicast routing protocol, typically PIM-SM, to get it to work over routed networks, but if you get it set up, you don’t need to create all VXLAN connections by hand.
The underlying encapsulation protocol for VXLAN is UDP.
Note: the id option is the VXLAN Network Identifier (VNI).
After that you need to bring the link up and either bridge it with another interface or assign an address to it.
GENEVE (Generic Virtual Network Encapsulation) is tunneling protocol that generally carries L2 frames. Its underlying encapsulation is UDP (the default port is 6081).
Its goals are similar to VXLAN, NVGRE, and STT; but unlike those protocols it’s extensible and supports adding arbitrary options to the header, to ensure it can adapt to future networking needs.
Like VXLAN, GENEVE uses 24-bit VNIs (virtual network identifiers). Per protocol specification, it can be either unicast or multicast, but the Linux kernel only supports the unicast mode for “internally” managed GENEVE tunnels as of Linux 6.10.
It’s possible to specify a custom UDP port instead of the default value of 6081:
It’s possible to create a GENEVE device without any addresses or anything else, and then use tc filters to encapsulate packets in GENEVE and add arbitrary headers.
Note: there can be only one externally managed GENEVE device per UDP port, so if you need multiple devices, use different dstport options for each one of them.
For IPv4 routes, you can use either a prefix length or a dotted-decimal subnet mask. That is, both 192.0.2.0/24 and 192.0.2.0/255.255.255.0 are equally acceptable.
The Linux kernel does not keep routes with unreachable next hops. If a link goes down, all routes that would use that link are permanently removed from the routing table. You might not have noticed that behaviour because, in many cases, additional software (e.g., NetworkManager or rp-pppoe) takes care of restoring routes when links go up and down.
If you are going to use your Linux machine as a router, consider installing a routing protocol suite such as FreeRangeRouting or BIRD. They keep track of link states and restore routes when a link goes up after going down. Of course, they also allow you to use dynamic routing protocols such as OSPF and BGP.
Some routes appear in the system without explicit configuration (“against your will”).
Once you assign an address to an interface, the system calculates its network address and creates a route to that network (this is why the subnet mask is required). Such routes are called connected routes.
For example, if you assign 203.0.113.25/24 to eth0, a connected route to 203.0.113.0/24 network will be created, and the system will know that hosts from that network can be reached directly.
When an interface goes down, connected routes associated with it are removed and all routes whose gateway belongs to the now inaccessible networks are removed as well because they fail gateway reachability check. The same mechanism prevents you from creating routes through inaccessible gateways.
You can use -4 and -6 options to view only IPv4 or IPv6 routes. By default, only IPv4 routes are displayed. To view IPv6 routes, use ip -6 route.
For example, if you use 192.168.0.0/24 subnet in your network and it’s broken into 192.168.0.0/25 and 192.168.0.128/25, you can see all those routes with ip route show to root 192.168.0.0/24.
Note: the word “to” is optional in all “show” commands.
If you want to view routes to 192.168.0.0/24 and all larger subnets, use ip route show to match 192.168.0.0/24.
Routers prefer more specific routes to less specific, so this is often useful for debugging in situations when traffic to a specific subnet is sent the wrong way because a route to it is missing, but routes to larger subnets exist.
If you want to see the routes to 192.168.0.0/24, but not to, say 192.168.0.0/25 and 192.168.0.0/16, you can use ip route show to exact 192.168.0.0/24.
Example: ip route get 192.168.0.0/24.
Note that this command always returns exactly one route. In most cases it’s not a problem, but in multi-path routing setups the result may not reflect the complete picture, so don’t forget to look at the corresponding “show” command output as well.
Until version 3.6, Linux used route caching. In older kernels, this command displays the contents of the route cache. It can be used with modifiers described above. In newer kernels, it does nothing.
Example: ip route add 192.0.2.0/25 dev ppp0
Interface routes are commonly used with point-to-point interfaces like PPP tunnels. Since there is no chance that more than one host is connected to a point-to-point interface, there’s also no need to specify the gateway address explicitly in those cases.
Example: ip route add 192.0.2.128/25 via 203.0.113.1 dev eth0 onlink
The onlink keyword disables the kernel’s gateway consistency checks and allows adding routes via gateways that look unreachable. Useful in tunnel configurations and container/virtualization networking where multiple networks on the same link use a single gateway.
For modifying routes, there are ip route change and ip route replace commands. The difference between them is that the change command will produce an error if you try to change a route that doesn’t exist. The replace command will create a route if it doesn’t exist already.
There is a shortcut for creating default routes.
For IPv6 routes, default is equivalent to ::/0.
Examples: ip route add blackhole 192.0.2.1/32.
Traffic to destinations that match a blackhole route is silently discarded.
There are two use cases for blackhole routes. First, they can work as a very fast outbound traffic filter, e.g., to make known botnet controllers inaccessible or to protect a server inside your network from an incoming DDoS attack. Second, they can be used to trick a routing protocol daemon into thinking that you have a route to a network if you only have real routes to its parts, but want to advertise it aggregated.
There are a few other types of special purpose routes.
These routes make the system discard packets and reply with an ICMP error message to the sender.
Unlike blackhole routes, these can’t be recommended for stopping unwanted traffic (e.g., DDoS) because they generate a reply packet for every discarded packet and thus create an even greater traffic flow. They can be good for implementing internal access policies, but a firewall is usually a better idea.
“Throw” routes may be used for implementing policy-based routing. In non-default tables they stop the lookup process but don’t send ICMP error messages.
If there are several routes to the same network with different metric value, the kernel prefers the one with the lowest metric.
An important part of this concept is that when an interface goes down, routes that would be rendered useless by this event disappear from the routing table (see the connected routes section), and the system will fall back to routes with higher metric values.
This feature is commonly used to implement backup connections to important destinations.
Multipath routes make the system balance packets across several links according to the weight (higher weight is preferred, so gateway/interface with weight 2 will get roughly two times more traffic than another one with weight 1). You can have as many gateways as you want, and you can mix gateway and interface routes:
Warning: the downside of this type of load balancing is that packets are not guaranteed to be sent back through the same link they came in. This is called “asymmetric routing”. For routers that simply forward packets and don’t do any local traffic processing, this is usually fine, and in some cases even unavoidable.
If your system does some local processing (e.g. NAT), this may cause problems with incoming connections. In that case, you should be using a stateful L4 load balancing setup instead.
Policy-based routing (PBR) in Linux is designed the following way: first you create custom routing tables, then you create rules to tell the kernel which tables to use for which packets.
User-defined tables are created automatically when you add the first route to them.
Note: You can use any route options described in the routing tables section for policy routes too, the only difference is the table ${table id/name} part at the end.
Numeric table identifiers and names can be used interchangeably. To create your own symbolic names, edit the /etc/iproute2/rt_tables config file.
delete, change, replace, and all other route actions work with any table too.
ip route … table main or ip route … table 254 have the exact same effect as commands without a table part.
Note: in this case, you need the show word; a shorthand like ip route table 120 does not work because the command would be ambiguous.
Traffic that matches the ${options} (described below) will be routed according to the table with specified name/id instead of the “main”/254 table if the lookup action is used.
blackhole, prohibit, and unreachable actions work just like in the default table.
For IPv6 rules, use ip -6, the rest of the syntax is the same.
table ${table id or name} can be used as a shortcut for lookup ${table id or name}.
Notes: all keyword can be used as a shortcut for 0.0.0.0/0 or ::/0
Examples: ip rule add tos 0x10 lookup 110.
Examples: ip rule add fwmark 0x11 lookup 100.
Make sure to set the mark in a firewall chain that is processed before the routing decision, else your PBR rules that use that mark will have no effect. You can find an excellent netfilter flowchart in Phil Hagen’s blog. For forwarded traffic, mangle FORWARD should be a good place, e.g. iptables -t mangle -I FORWARD -s 192.0.2.1 -j MARK –set-mark 0x11.
Rules with iif lo (loopback) will match locally generated traffic.
Examples: ip rule add oif eth0 lookup 10.
Note: this works only for locally generated traffic.
To apply a rule to a single user, use the same UID for both the start and end of the range.
Note: Rules are traversed from the lowest to the highest priority, and processing stops on the first match, so you need to put more specific rules before less specific ones. The example above demonstrates rules for 192.0.2.0/24 and its subnet 192.0.2.0/25. If the priorities were reversed and the rule for /25 was placed after the rule for /24, it would never be reached.
Examples: ip rule del 192.0.2.0/24 lookup 10
Notes: You can copy/paste from the output of ip rule show or ip -6 rule show.
Notes: this operation is highly disruptive. Even if you have not configured any rules, some fundamental rules like from all lookup main rules are created for you by default. On an unconfigured machine, you can see this:
The from all lookup local rule is special and cannot be deleted. The from all lookup main is not, there may be valid reasons not to have it, e.g., if you want to route only traffic you created explicit rules for. As a side effect, if you do ip rule flush, that rule will be deleted, which will make the system stop routing any traffic until you restore your rules.
VRF (Virtual Routing and Forwarding) is a mechanism for isolating routes of a network in a separate routing table. It allows multiple networks with conflicting address ranges to co-exist in the same router. The most common use case for it is multi-tenant setups and provider-supported VPNs where customers can use their own network addresses.
The main difference from policy-based routing is that “normal” non-default tables used in PBR only separate static routes but not connected routes, so they cannot resolve address conflicts. When a network interface is bound to a VRF, all connected routes from it will be moved to a separate routing table.
Unlike network namespaces, VRFs work exclusively on the network layer. They do not create a separate copy of the network stack, do not interfere with L2 protocols such as LLDP, and one process can bind sockets to multiple VRFs (useful for dynamic routing protocol or IPsec daemons).
Example: sudo ip link add foo type vrf table 100
If routing table 100 does not exist, it will be automatically created.
The output of ip vrf show only shows associations between VRFs and routing tables, but not associations between VRFs and network interfaces. To view all interfaces of a VRF, use ip link show vrf ${vrf}.
Example: ip link set eth0 master foo.
All connected routes associated with the interface will be moved to the VRF table. For example, if eth0 has address 192.0.2.1/24, and a VRF instance foo uses table 100, then the route to 192.0.2.0/24 will disappear from the main table and re-appear in table 100.
Connected routes associated with that interface will be moved back to the main table.
Example: ip vrf exec ping 192.0.2.100.
The traffic of the process will be routed according to the VRF table routes, so it’s useful for troubleshooting.
If the process with PID ${pid} is running in a VRF, the command will display its name. If it is not running in a VRF, the command output is empty.
If ${vrf} does not exist, the command will show an error message. If it exists but there are no processed running in it, the output is empty.
Network namespaces are isolated network stack instances within a single machine. They can be used for security domain separation, managing traffic flows between virtual machines, and so on.
Every namespace is a complete copy of the networking stack with its own interfaces, addresses, routes etc. You can run processes inside a namespace and bridge namespaces to physical interfaces.
Example: ip netns exec foo /bin/sh.
Note: assigning a process to a non-default namespace requires root privileges.
You can run any processes inside a namespace, including an interactive shell. If you want to run ip itself in another namespace, you can use either ip netns exec ${namespace name} ip ${ip subcommand} (same as with any other command), or the -n/-netns shortcut: ip -n ${namespace name} ${ip subcommand}.
The output will be a list of PIDs.
Example: ip netns identify 9000.
Example: ip link set dev eth0.100 netns foo.
Note: once you assign an interface to a namespace, it disappears from the default namespace, and you will have to perform all operations with it via ip netns exec ${namespace name} or ip -n ${namespace name} ${ip subcommand}, like ip -n ${namespace name} link set dev dummy0 down.
Moreover, when you move an interface to another namespace, it loses all existing configuration such as IP addresses configured on it and goes to the DOWN state. You’ll need to bring it back up and reconfigure it.
If you specify a PID instead of a namespace name, the interface gets assigned to the primary namespace of the process with that PID. This way you can reassign an interface back to the default namespace with e.g., ip -n ${namespace name} link set dev ${intf} netns 1 (since init or another process with PID 1 is pretty much guaranteed to be in default namespace).
This can be done by creating a pair of veth links and assigning them to different namespaces.
Suppose you want to connect a namespace named “foo” to the default namespace. First, create a pair of veth devices: ip link add name veth1 type veth peer name veth2.
Move veth2 to namespace foo: ip link set dev veth2 netns foo.
Bring veth2 up and add an address in “foo” namespace:
Add an address to veth1, which stays in the default namespace: ip address add 10.1.1.2/24 dev veth1.
Now you can ping 10.1.1.1, which is in the foo namespace, and set up routes to subnets configured in other interfaces of that namespace.
If you want switching instead of routing, you can bridge those veth interfaces with other interfaces in corresponding namespaces. The same technique can be used for connecting namespaces to physical networks.
Displays events such as creation and deletion of namespaces when they occur.
Multicast is mostly handled by applications and routing daemons, so there is not much you can and should do manually here. Multicast-related ip commands are mostly useful for debugging.
You cannot join an IP multicast group manually, but you can add a multicast MAC address (even though it’s rarely needed).
Example: ip maddress add 01:00:5e:00:00:ab dev eth0.
Multicast routes cannot be added manually, so this command can only show multicast routes installed by a routing daemon. It supports the same modifiers as unicast route viewing commands (iif, table, from etc.).
You can monitor certain network events with iproute2, such as changes in network configuration, routing tables, and ARP/NDP tables.
You may either call the command without parameters or explicitly specify all.
When there are distinct IPv4 and IPv6 subsystems, the usual -4 and -6 options allow you to display events only for the specified protocol.
iproute2 includes a program called rtmon that serves essentially the same purpose but writes events to a binary log file instead of displaying them. You can read those log files with the ip monitor command:
The rtmon syntax is similar to that of ip monitor, except event types are limited to link, address, route, and all; and address family is specified using the -family option:
Example: ip netconf show dev eth0
Content: Nicolas Dichtel, Russel Stuart, Phil Huang, Haishan, Emil Pederson, Nick B.
Grammar, style, typo fixes: Trick van Staveren, powyginanachochla, Nathan Handler, Bhaskar Sarma Upadhyayula, Geert Stappers, Alex White-Robinson, Achilleas Pipinellis, fauxm, fgtham, eri, Zhuoyun Wei, Jonathan ZHAO, Julien Barbot, thoastbrot.
Source: Hacker News — This article was automatically imported from the source. Read full article at original source →
