IPRoute module

Classes

The RTNL API is provided by the class RTNL_API. It is a mixin class that works on top of any RTNL-compatible socket, so several classes with almost the same API are available:

  • IPRoute -- simple RTNL API

  • NetNS -- RTNL API in a network namespace

  • IPBatch -- RTNL packet compiler

  • RemoteIPRoute -- run RTNL remotely (no deployment required)

Responses as lists

The netlink socket implementation in the pyroute2 is agnostic to particular netlink protocols, and always returns a list of messages as the response to a request sent to the kernel:

with IPRoute() as ipr:

    # this request returns one match
    eth0 = ipr.link_lookup(ifname='eth0')
    len(eth0)  # -> 1, if exists, else 0

    # but that one returns a set of
    up = ipr.link_lookup(operstate='UP')
    len(up)  # -> k, where 0 <= k <= [interface count]

Thus, always expect a list in the response, running any IPRoute() netlink request.

NLMSG_ERROR responses

Some kernel subsystems return NLMSG_ERROR in response to any request. It is OK as long as nlmsg["header"]["error"] is None. Otherwise an exception will be raised by the parser.

So if instead of an exception you get a NLMSG_ERROR message, it means error == 0, the same as $? == 0 in bash.

How to work with messages

Every netlink message contains header, fields and NLAs (netlink attributes). Every NLA is a netlink message... (see "recursion").

And the library provides parsed messages according to this scheme. Every RTNL message contains:

  • nlmsg['header'] -- parsed header

  • nlmsg['attrs'] -- NLA chain (parsed on demand)

  • 0 .. k data fields, e.g. nlmsg['flags'] etc.

  • nlmsg.header -- the header fields spec

  • nlmsg.fields -- the data fields spec

  • nlmsg.nla_map -- NLA spec

An important parser feature is that NLAs are parsed on demand, when someone tries to access them. Otherwise the parser doesn't waste CPU cycles.

The NLA chain is a list-like structure, not a dictionary. The netlink standard doesn't require NLAs to be unique within one message:

{'attrs': [('IFLA_IFNAME', 'lo'),    # [1]
           ('IFLA_TXQLEN', 1),
           ('IFLA_OPERSTATE', 'UNKNOWN'),
           ('IFLA_LINKMODE', 0),
           ('IFLA_MTU', 65536),
           ('IFLA_GROUP', 0),
           ('IFLA_PROMISCUITY', 0),
           ('IFLA_NUM_TX_QUEUES', 1),
           ('IFLA_NUM_RX_QUEUES', 1),
           ('IFLA_CARRIER', 1),
           ...],
 'change': 0,
 'event': 'RTM_NEWLINK',             # [2]
 'family': 0,
 'flags': 65609,
 'header': {'error': None,           # [3]
            'flags': 2,
            'length': 1180,
            'pid': 28233,
            'sequence_number': 257,  # [4]
            'type': 16},             # [5]
 'ifi_type': 772,
 'index': 1}

 # [1] every NLA is parsed upon access
 # [2] this field is injected by the RTNL parser
 # [3] if not None, an exception will be raised
 # [4] more details in the netlink description
 # [5] 16 == RTM_NEWLINK

To access fields:

msg['index'] == 1

To access one NLA:

msg.get_attr('IFLA_CARRIER') == 1

When an NLA with the specified name is not present in the chain, get_attr() returns None. To get the list of all NLAs of that name, use get_attrs(). A real example with NLA hierarchy, take notice of get_attr() and get_attrs() usage:

# for macvlan interfaces there may be several
# IFLA_MACVLAN_MACADDR NLA provided, so use
# get_attrs() to get all the list, not only
# the first one

(msg
 .get_attr('IFLA_LINKINFO')           # one NLA
 .get_attr('IFLA_INFO_DATA')          # one NLA
 .get_attrs('IFLA_MACVLAN_MACADDR'))  # a list of

The protocol itself has no limit for number of NLAs of the same type in one message, that's why we can not make a dictionary from them -- unlike PF_ROUTE messages.

BSD systems

The library provides very basic RTNL API for BSD systems via protocol emulation. Only getters are supported yet, no setters.

BSD employs PF_ROUTE sockets to send notifications about network object changes, but the protocol doesn not allow changing links/addresses/etc like Netlink.

To change network setting one have to rely on system calls or external tools. Thus IPRoute on BSD systems is not as effective as on Linux, where all the changes are done via Netlink.

The monitoring started with bind() is implemented as an implicit thread, started by the bind() call. This is done to have only one notification FD, used both for normal calls and notifications. This allows to use IPRoute objects in poll/select calls.

On Linux systems RTNL API is provided by the netlink protocol, so no implicit threads are started by default to monitor the system updates. IPRoute.bind(...) may start the async cache thread, but only when asked explicitly:

#
# Normal monitoring. Always starts monitoring thread on
# FreeBSD / OpenBSD, no threads on Linux.
#
with IPRoute() as ipr:
    ipr.bind()
    ...

#
# Monitoring with async cache. Always starts cache thread
# on Linux, ignored on FreeBSD / OpenBSD.
#
with IPRoute() as ipr:
    ipr.bind(async_cache=True)
    ...

On all the supported platforms, be it Linux or BSD, the IPRoute.recv(...) method returns valid netlink RTNL raw binary payload and IPRoute.get(...) returns parsed RTNL messages.

Windows systems

Windows systems are not supported, but the library provides some proof-of-concept how to build an RTNL-compatible core on top of WinAPI calls.

Only two methods are provided so far. If you're interested in extending the functionality, you're welcome to propose PRs.

Warning

Using pyroute2 on Windows requires installing win_inet_pton module, you can use pip install win_inet_pton.

class pyroute2.iproute.windows.IPRoute(*argv, **kwarg)

Get network interfaces list:

>>> pprint(ipr.get_links())
[{'attrs': (['IFLA_ADDRESS', '52:54:00:7a:8a:49'],
            ['IFLA_IFNAME',
             '{F444467B-3549-455D-81F2-AB617C7421AB}']),
  'change': 0,
  'family': 0,
  'flags': 0,
  'header': {},
  'ifi_type': 0,
  'index': 7}]
get_addr(*argv, **kwarg)

Get IP addresses:

>>> pprint(ipr.get_addr())
[{'attrs': (['IFA_ADDRESS', '192.168.122.81'],
            ['IFA_LOCAL', '192.168.122.81'],
            ['IFA_LABEL',
             '{F444467B-3549-455D-81F2-AB617C7421AB}']),
  'family': <AddressFamily.AF_INET: 2>,
  'flags': 0,
  'header': {},
  'index': 7,
  'prefixlen': 24,
  'scope': 0}]

Linux systems

class pyroute2.iproute.linux.RTNL_API(*argv, **kwarg)

RTNL_API should not be instantiated by itself. It is intended to be used as a mixin class. Following classes use RTNL_API:

  • IPRoute -- RTNL API to the current network namespace

  • NetNS -- RTNL API to another network namespace

  • IPBatch -- RTNL compiler

  • ShellIPR -- RTNL via standard I/O, runs IPRoute in a shell

It is an old-school API, that provides access to rtnetlink as is. It helps you to retrieve and change almost all the data, available through rtnetlink:

from pyroute2 import IPRoute
ipr = IPRoute()
# create an interface
ipr.link('add', ifname='brx', kind='bridge')
# lookup the index
dev = ipr.link_lookup(ifname='brx')[0]
# bring it down
ipr.link('set', index=dev, state='down')
# change the interface MAC address and rename it just for fun
ipr.link('set', index=dev,
         address='00:11:22:33:44:55',
         ifname='br-ctrl')
# add primary IP address
ipr.addr('add', index=dev,
         address='10.0.0.1', mask=24,
         broadcast='10.0.0.255')
# add secondary IP address
ipr.addr('add', index=dev,
         address='10.0.0.2', mask=24,
         broadcast='10.0.0.255')
# bring it up
ipr.link('set', index=dev, state='up')
filter_messages(dump_filter, msgs)

Filter messages using dump_filter. The filter might be a callable, then it will be called for every message in the list. Or it might be a dict, where keys are used to get values from messages, and dict values are used to match the message.

The method might be called directly. It is also used by calls like ipr.link('dump', ....), where keyword arguments work as dump_filter for ipr.filter_messages().

A callable dump_filter must return True or False:

# get all links with names starting with eth:
#
ipr.filter_messages(
    lambda x: x.get_attr('IFLA_IFNAME').startswith('eth'),
    ipr.link('dump')
)

A dict dump_filter can have callables as values:

# get all links with names starting with eth, and
# MAC address in a database:
#
ipr.filter_messages(
    {
        'ifname': lambda x: x.startswith('eth'),
        'address': lambda x: x in database,
    },
    ipr.link('dump')
)

... or constants to compare with:

# get all links in state up:
#
ipr.filter_message({'state': 'up'}, ipr.link('dump'))
dump(groups=None)

Dump network objects.

On OpenBSD:

  • get_links()

  • get_addr()

  • get_neighbours()

  • get_routes()

On Linux:

  • get_links()

  • get_addr()

  • get_neighbours()

  • get_vlans()

  • dump FDB

  • IPv4 and IPv6 rules

poll(method, command, timeout=10, interval=0.2, **spec)

Run method with a positional argument command and keyword arguments **spec every interval seconds, but not more than timeout, until it returns a result which doesn't evaluate to False.

Example:

# create a bridge interface and wait for it:
#
spec = {
    'ifname': 'br0',
    'kind': 'bridge',
    'state': 'up',
    'br_stp_state': 1,
}
ipr.link('add', **spec)
ret = ipr.poll(ipr.link, 'dump', **spec)

assert ret[0].get('ifname') == 'br0'
assert ret[0].get('state') == 'up'
assert ret[0].get(('linkinfo', 'data', 'br_stp_state')) == 1
get_qdiscs(index=None)

Get all queue disciplines for all interfaces or for specified one.

get_filters(index=0, handle=0, parent=0)

Get filters for specified interface, handle and parent.

get_classes(index=0)

Get classes for specified interface.

get_vlans(**kwarg)

Dump available vlan info on bridge ports

Get network interfaces.

By default returns all interfaces. Arguments vector can contain interface indices or a special keyword 'all':

ip.get_links()
ip.get_links('all')
ip.get_links(1, 2, 3)

interfaces = [1, 2, 3]
ip.get_links(*interfaces)
get_neighbours(family=AddressFamily.AF_UNSPEC, match=None, **kwarg)

Dump ARP cache records.

The family keyword sets the family for the request: e.g. AF_INET or AF_INET6 for arp cache, AF_BRIDGE for fdb.

If other keyword arguments not empty, they are used as filter. Also, one can explicitly set filter as a function with the match parameter.

Examples:

# get neighbours on the 3rd link:
ip.get_neighbours(ifindex=3)

# get a particular record by dst:
ip.get_neighbours(dst='172.16.0.1')

# get fdb records:
ip.get_neighbours(AF_BRIDGE)

# and filter them by a function:
ip.get_neighbours(AF_BRIDGE, match=lambda x: x['state'] == 2)
get_ntables(family=AddressFamily.AF_UNSPEC)

Get neighbour tables

get_addr(family=AddressFamily.AF_UNSPEC, match=None, **kwarg)

Dump addresses.

If family is not specified, both AF_INET and AF_INET6 addresses will be dumped:

# get all addresses
ip.get_addr()

It is possible to apply filters on the results:

# get addresses for the 2nd interface
ip.get_addr(index=2)

# get addresses with IFA_LABEL == 'eth0'
ip.get_addr(label='eth0')

# get all the subnet addresses on the interface, identified
# by broadcast address (should be explicitly specified upon
# creation)
ip.get_addr(index=2, broadcast='192.168.1.255')

A custom predicate can be used as a filter:

ip.get_addr(match=lambda x: x['index'] == 1)
get_rules(family=AddressFamily.AF_UNSPEC, match=None, **kwarg)

Get all rules. By default return all rules. To explicitly request the IPv4 rules use family=AF_INET.

Example::

ip.get_rules() # get all the rules for all families ip.get_rules(family=AF_INET6) # get only IPv6 rules

get_routes(family=255, match=None, **kwarg)

Get all routes. You can specify the table. There are up to 4294967295 routing classes (tables), and the kernel returns all the routes on each request. So the routine filters routes from full output. Note the number of tables is increased from 255 in Linux 2.6+.

Example:

ip.get_routes()  # get all the routes for all families
ip.get_routes(family=AF_INET6)  # get only IPv6 routes
ip.get_routes(table=254)  # get routes from 254 table

The default family=255 is a hack. Despite the specs, the kernel returns only IPv4 routes for AF_UNSPEC family. But it returns all the routes for all the families if one uses an invalid value here. Hack but true. And let's hope the kernel team will not fix this bug.

static open_file(path)

Open a file (read only) and return its (fd, inode).

static close_file(fd)

Close a file that was previously opened with open_file().

static get_pid()

Return the PID of the current process.

get_netnsid(nsid=None, pid=None, fd=None, target_nsid=None)

Return a dict containing the result of a RTM_GETNSID query. This loosely corresponds to the "ip netns list-id" command.

get_netns_info(list_proc=False)

A prototype method to list available netns and associated interfaces. A bit weird to have it here and not under pyroute2.netns, but it uses RTNL to get all the info.

get_default_routes(family=AddressFamily.AF_UNSPEC, table=254)

Get default routes

Lookup interface index (indeces) by first level NLA value.

Example:

ip.link_lookup(address="52:54:00:9d:4e:3d")
ip.link_lookup(ifname="lo")
ip.link_lookup(operstate="UP")

Please note, that link_lookup() returns list, not one value.

flush_routes(*argv, **kwarg)

Flush routes -- purge route records from a table. Arguments are the same as for get_routes() routine. Actually, this routine implements a pipe from get_routes() to nlm_request().

flush_addr(*argv, **kwarg)

Flush IP addresses.

Examples:

# flush all addresses on the interface with index 2:
ipr.flush_addr(index=2)

# flush all addresses with IFA_LABEL='eth0':
ipr.flush_addr(label='eth0')
flush_rules(*argv, **kwarg)

Flush rules. Please keep in mind, that by default the function operates on all rules of all families. To work only on IPv4 rules, one should explicitly specify family=AF_INET.

Examples:

# flush all IPv4 rule with priorities above 5 and below 32000
ipr.flush_rules(family=AF_INET, priority=lambda x: 5 < x < 32000)

# flush all IPv6 rules that point to table 250:
ipr.flush_rules(family=socket.AF_INET6, table=250)
brport(command, **kwarg)

Set bridge port parameters. Example:

idx = ip.link_lookup(ifname='eth0')
ip.brport("set", index=idx, unicast_flood=0, cost=200)
ip.brport("show", index=idx)

Possible keywords are NLA names for the protinfo_bridge class, without the prefix and in lower letters.

vlan_filter(command, **kwarg)

Vlan filters is another approach to support vlans in Linux. Before vlan filters were introduced, there was only one way to bridge vlans: one had to create vlan interfaces and then add them as ports:

        +------+      +----------+
net --> | eth0 | <--> | eth0.500 | <---+
        +------+      +----------+     |
                                       v
        +------+                    +-----+
net --> | eth1 |                    | br0 |
        +------+                    +-----+
                                       ^
        +------+      +----------+     |
net --> | eth2 | <--> | eth2.500 | <---+
        +------+      +----------+

It means that one has to create as many bridges, as there were vlans. Vlan filters allow to bridge together underlying interfaces and create vlans already on the bridge:

# v500 label shows which interfaces have vlan filter

        +------+ v500
net --> | eth0 | <-------+
        +------+         |
                         v
        +------+      +-----+    +---------+
net --> | eth1 | <--> | br0 |<-->| br0v500 |
        +------+      +-----+    +---------+
                         ^
        +------+ v500    |
net --> | eth2 | <-------+
        +------+

In this example vlan 500 will be allowed only on ports eth0 and eth2, though all three eth nics are bridged.

Some example code:

# create bridge
ip.link("add",
        ifname="br0",
        kind="bridge")

# attach a port
ip.link("set",
        index=ip.link_lookup(ifname="eth0")[0],
        master=ip.link_lookup(ifname="br0")[0])

# set vlan filter
ip.vlan_filter("add",
               index=ip.link_lookup(ifname="eth0")[0],
               vlan_info={"vid": 500})

# create vlan interface on the bridge
ip.link("add",
        ifname="br0v500",
        kind="vlan",
        link=ip.link_lookup(ifname="br0")[0],
        vlan_id=500)

# set all UP
ip.link("set",
        index=ip.link_lookup(ifname="br0")[0],
        state="up")
ip.link("set",
        index=ip.link_lookup(ifname="br0v500")[0],
        state="up")
ip.link("set",
        index=ip.link_lookup(ifname="eth0")[0],
        state="up")

# set IP address
ip.addr("add",
        index=ip.link_lookup(ifname="br0v500")[0],
        address="172.16.5.2",
        mask=24)

Now all the traffic to the network 172.16.5.2/24 will go
to vlan 500 only via ports that have such vlan filter.

Required arguments for vlan_filter() -- index and vlan_info. Vlan info struct:

{"vid": uint16,
 "flags": uint16}
More details:
  • kernel:Documentation/networking/switchdev.txt

  • pyroute2.netlink.rtnl.ifinfmsg:... vlan_info

One can specify flags as int or as a list of flag names:
  • master == 0x1

  • pvid == 0x2

  • untagged == 0x4

  • range_begin == 0x8

  • range_end == 0x10

  • brentry == 0x20

E.g.:

{"vid": 20,
 "flags": ["pvid", "untagged"]}

# is equal to
{"vid": 20,
 "flags": 6}

Commands:

add

Add vlan filter to a bridge port. Example:

ip.vlan_filter("add", index=2, vlan_info={"vid": 200})

del

Remove vlan filter from a bridge port. Example:

ip.vlan_filter("del", index=2, vlan_info={"vid": 200})
fdb(command, **kwarg)

Bridge forwarding database management.

More details:
  • kernel:Documentation/networking/switchdev.txt

  • pyroute2.netlink.rtnl.ndmsg

add

Add a new FDB record. Works in the same way as ARP cache management, but some additional NLAs can be used:

# simple FDB record
#
ip.fdb('add',
       ifindex=ip.link_lookup(ifname='br0')[0],
       lladdr='00:11:22:33:44:55',
       dst='10.0.0.1')

# specify vlan
# NB: vlan should exist on the device, use
# `vlan_filter()`
#
ip.fdb('add',
       ifindex=ip.link_lookup(ifname='br0')[0],
       lladdr='00:11:22:33:44:55',
       dst='10.0.0.1',
       vlan=200)

# specify vxlan id and port
# NB: works only for vxlan devices, use
# `link("add", kind="vxlan", ...)`
#
# if port is not specified, the default one is used
# by the kernel.
#
# if vni (vxlan id) is equal to the device vni,
# the kernel doesn't report it back
#
ip.fdb('add',
       ifindex=ip.link_lookup(ifname='vx500')[0]
       lladdr='00:11:22:33:44:55',
       dst='10.0.0.1',
       port=5678,
       vni=600)

append

Append a new FDB record. The same syntax as for add.

del

Remove an existing FDB record. The same syntax as for add.

dump

Dump all the FDB records. If any **kwarg is provided, results will be filtered:

# dump all the records
ip.fdb('dump')

# show only specific lladdr, dst, vlan etc.
ip.fdb('dump', lladdr='00:11:22:33:44:55')
ip.fdb('dump', dst='10.0.0.1')
ip.fdb('dump', vlan=200)
neigh(command, **kwarg)

Neighbours operations, same as ip neigh or bridge fdb

add

Add a neighbour record, e.g.:

from pyroute2 import IPRoute
from pyroute2.netlink.rtnl import ndmsg

# add a permanent record on veth0
idx = ip.link_lookup(ifname='veth0')[0]
ip.neigh('add',
         dst='172.16.45.1',
         lladdr='00:11:22:33:44:55',
         ifindex=idx,
         state=ndmsg.states['permanent'])

set

Set an existing record or create a new one, if it doesn't exist. The same as above, but the command is "set":

ip.neigh('set',
         dst='172.16.45.1',
         lladdr='00:11:22:33:44:55',
         ifindex=idx,
         state=ndmsg.states['permanent'])

change

Change an existing record. If the record doesn't exist, fail.

del

Delete an existing record.

dump

Dump all the records in the NDB:

ip.neigh('dump')

get

Get specific record (dst and ifindex are mandatory). Available only on recent kernel:

ip.neigh('get',
         dst='172.16.45.1',
         ifindex=idx)

Link operations.

Keywords to set up ifinfmsg fields:
  • index -- interface index

  • family -- AF_BRIDGE for bridge operations, otherwise 0

  • flags -- device flags

  • change -- change mask

All other keywords will be translated to NLA names, e.g. mtu -> IFLA_MTU, af_spec -> IFLA_AF_SPEC etc. You can provide a complete NLA structure or let filters do it for you. E.g., these pairs show equal statements:

# set device MTU
ip.link("set", index=x, mtu=1000)
ip.link("set", index=x, IFLA_MTU=1000)

# add vlan device
ip.link("add", ifname="test", kind="dummy")
ip.link("add", ifname="test",
        IFLA_LINKINFO={'attrs': [['IFLA_INFO_KIND', 'dummy']]})

Filters are implemented in the pyroute2.iproute.req module. You can contribute your own if you miss shortcuts.

Commands:

add

To create an interface, one should specify the interface kind:

ip.link("add",
        ifname="test",
        kind="dummy")

The kind can be any of those supported by kernel. It can be dummy, bridge, bond etc. On modern kernels one can specify even interface index:

ip.link("add",
        ifname="br-test",
        kind="bridge",
        index=2345)

Specific type notes:

► geneve

Create GENEVE tunnel:

ip.link("add",
        ifname="genx",
        kind="geneve",
        geneve_id=42,
        geneve_remote="172.16.0.101")

Support for GENEVE over IPv6 is also included; use geneve_remote6 to configure a remote IPv6 address.

► gre

Create GRE tunnel:

ip.link("add",
        ifname="grex",
        kind="gre",
        gre_local="172.16.0.1",
        gre_remote="172.16.0.101",
        gre_ttl=16)

The keyed GRE requires explicit iflags/oflags specification:

ip.link("add",
        ifname="grex",
        kind="gre",
        gre_local="172.16.0.1",
        gre_remote="172.16.0.101",
        gre_ttl=16,
        gre_ikey=10,
        gre_okey=10,
        gre_iflags=32,
        gre_oflags=32)

Support for GRE over IPv6 is also included; use kind=ip6gre and ip6gre_ as the prefix for its values.

► ipip

Create ipip tunnel:

ip.link("add",
        ifname="tun1",
        kind="ipip",
        ipip_local="172.16.0.1",
        ipip_remote="172.16.0.101",
        ipip_ttl=16)

Support for sit and ip6tnl is also included; use kind=sit and sit_ as prefix for sit tunnels, and kind=ip6tnl and ip6tnl_ prefix for ip6tnl tunnels.

► macvlan

Macvlan interfaces act like VLANs within OS. The macvlan driver provides an ability to add several MAC addresses on one interface, where every MAC address is reflected with a virtual interface in the system.

In some setups macvlan interfaces can replace bridge interfaces, providing more simple and at the same time high-performance solution:

ip.link("add",
        ifname="mvlan0",
        kind="macvlan",
        link=ip.link_lookup(ifname="em1")[0],
        macvlan_mode="private").commit()

Several macvlan modes are available: "private", "vepa", "bridge", "passthru". Ususally the default is "vepa".

► macvtap

Almost the same as macvlan, but creates also a character tap device:

ip.link("add",
        ifname="mvtap0",
        kind="macvtap",
        link=ip.link_lookup(ifname="em1")[0],
        macvtap_mode="vepa").commit()

Will create a device file "/dev/tap%s" % index

► tuntap

Possible tuntap keywords:

  • mode — "tun" or "tap"

  • uid — integer

  • gid — integer

  • ifr — dict of tuntap flags (see ifinfmsg:... tuntap_data)

Create a tap interface:

ip.link("add",
        ifname="tap0",
        kind="tuntap",
        mode="tap")

Tun/tap interfaces are created using ioctl(), but the library provides a transparent way to manage them using netlink API.

► veth

To properly create veth interface, one should specify peer also, since veth interfaces are created in pairs:

# simple call
ip.link("add", ifname="v1p0", kind="veth", peer="v1p1")

# set up specific veth peer attributes
ip.link("add",
        ifname="v1p0",
        kind="veth",
        peer={"ifname": "v1p1",
              "net_ns_fd": "test_netns"})

► vlan

VLAN interfaces require additional parameters, vlan_id and link, where link is a master interface to create VLAN on:

ip.link("add",
        ifname="v100",
        kind="vlan",
        link=ip.link_lookup(ifname="eth0")[0],
        vlan_id=100)

There is a possibility to create also 802.1ad interfaces:

# create external vlan 802.1ad, s-tag
ip.link("add",
        ifname="v100s",
        kind="vlan",
        link=ip.link_lookup(ifname="eth0")[0],
        vlan_id=100,
        vlan_protocol=0x88a8)

# create internal vlan 802.1q, c-tag
ip.link("add",
        ifname="v200c",
        kind="vlan",
        link=ip.link_lookup(ifname="v100s")[0],
        vlan_id=200,
        vlan_protocol=0x8100)

► vrf

VRF interfaces (see linux/Documentation/networking/vrf.txt):

ip.link("add",
        ifname="vrf-foo",
        kind="vrf",
        vrf_table=42)

► vxlan

VXLAN interfaces are like VLAN ones, but require a bit more parameters:

ip.link("add",
        ifname="vx101",
        kind="vxlan",
        vxlan_link=ip.link_lookup(ifname="eth0")[0],
        vxlan_id=101,
        vxlan_group='239.1.1.1',
        vxlan_ttl=16)

All possible vxlan parameters are listed in the module pyroute2.netlink.rtnl.ifinfmsg:... vxlan_data.

► ipoib

IPoIB driver provides an ability to create several ip interfaces on one interface. IPoIB interfaces requires the following parameter:

link : The master interface to create IPoIB on.

The following parameters can also be provided:

  • pkey- Inifiniband partition key the ip interface is associated with

  • mode- Underlying infiniband transport mode. One of: ['datagram' ,'connected']

  • umcast- If set(1), multicast group membership for this interface is handled by user space.

Example:

ip.link("add",
        ifname="ipoib1",
        kind="ipoib",
        link=ip.link_lookup(ifname="ib0")[0],
        pkey=10)

set

Set interface attributes:

# get interface index
x = ip.link_lookup(ifname="eth0")[0]
# put link down
ip.link("set", index=x, state="down")
# rename and set MAC addr
ip.link("set", index=x, address="00:11:22:33:44:55", name="bala")
# set MTU and TX queue length
ip.link("set", index=x, mtu=1000, txqlen=2000)
# bring link up
ip.link("set", index=x, state="up")

Seting bridge or tunnel attributes require kind to be specified in order to properly encode IFLA_LINKINFO:

ip.link("set",
        index=x,
        kind="bridge",
        br_forward_delay=2000)

ip.link("set",
        index=x,
        kind="gre",
        gre_local="10.0.0.1",
        gre_remote="10.1.0.103")

Keyword "state" is reserved. State can be "up" or "down", it is a shortcut:

state="up":   flags=1, mask=1
state="down": flags=0, mask=0

SR-IOV virtual function setup:

# get PF index
x = ip.link_lookup(ifname="eth0")[0]
# setup macaddr
ip.link("set",
        index=x,                          # PF index
        vf={"vf": 0,                      # VF index
            "mac": "00:11:22:33:44:55"})  # address
# setup vlan
ip.link("set",
        index=x,           # PF index
        vf={"vf": 0,       # VF index
            "vlan": 100})  # the simplest case
# setup QinQ
ip.link("set",
        index=x,                           # PF index
        vf={"vf": 0,                       # VF index
            "vlan": [{"vlan": 100,         # vlan id
                      "proto": 0x88a8},    # 802.1ad
                     {"vlan": 200,         # vlan id
                      "proto": 0x8100}]})  # 802.1q

update

Almost the same as set, except it uses different flags and message type. Mostly does the same, but in some cases differs. If you're not sure what to use, use set.

del

Destroy the interface:

ip.link("del", index=ip.link_lookup(ifname="dummy0")[0])

dump

Dump info for all interfaces

get

Get specific interface info:

ip.link("get", index=ip.link_lookup(ifname="br0")[0])

Get extended attributes like SR-IOV setup:

ip.link("get", index=3, ext_mask=1)
addr(command, *argv, **kwarg)

Address operations

  • command -- add, delete, replace, dump

  • index -- device index

  • address -- IPv4 or IPv6 address

  • mask -- address mask

  • family -- socket.AF_INET for IPv4 or socket.AF_INET6 for IPv6

  • scope -- the address scope, see /etc/iproute2/rt_scopes

  • kwarg -- dictionary, any ifaddrmsg field or NLA

Later the method signature will be changed to:

def addr(self, command, match=None, **kwarg):
    # the method body

So only keyword arguments (except of the command) will be accepted. The reason for this change is an unification of API.

Example:

idx = 62
ip.addr('add', index=idx, address='10.0.0.1', mask=24)
ip.addr('add', index=idx, address='10.0.0.2', mask=24)

With more NLAs:

# explicitly set broadcast address
ip.addr('add', index=idx,
        address='10.0.0.3',
        broadcast='10.0.0.255',
        prefixlen=24)

# make the secondary address visible to ifconfig: add label
ip.addr('add', index=idx,
        address='10.0.0.4',
        broadcast='10.0.0.255',
        prefixlen=24,
        label='eth0:1')

Configure p2p address on an interface:

ip.addr('add', index=idx,
        address='10.1.1.2',
        mask=24,
        local='10.1.1.1')
tc(command, kind=None, index=0, handle=0, **kwarg)

"Swiss knife" for traffic control. With the method you can add, delete or modify qdiscs, classes and filters.

  • command -- add or delete qdisc, class, filter.

  • kind -- a string identifier -- "sfq", "htb", "u32" and so on.

  • handle -- integer or string

Command can be one of ("add", "del", "add-class", "del-class", "add-filter", "del-filter") (see commands dict in the code).

Handle notice: traditional iproute2 notation, like "1:0", actually represents two parts in one four-bytes integer:

1:0    ->    0x10000
1:1    ->    0x10001
ff:0   ->   0xff0000
ffff:1 -> 0xffff0001

Target notice: if your target is a class/qdisc that applies an algorithm that can only apply to upstream traffic profile, but your keys variable explicitly references a match that is only relevant for upstream traffic, the kernel will reject the filter. Unless you're dealing with devices like IMQs

For pyroute2 tc() you can use both forms: integer like 0xffff0000 or string like 'ffff:0000'. By default, handle is 0, so you can add simple classless queues w/o need to specify handle. Ingress queue causes handle to be 0xffff0000.

So, to set up sfq queue on interface 1, the function call will be like that:

ip = IPRoute()
ip.tc("add", "sfq", 1)

Instead of string commands ("add", "del"...), you can use also module constants, RTM_NEWQDISC, RTM_DELQDISC and so on:

ip = IPRoute()
flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE | NLM_F_EXCL
ip.tc((RTM_NEWQDISC, flags), "sfq", 1)

It should be noted that "change", "change-class" and "change-filter" work like "replace", "replace-class" and "replace-filter", except they will fail if the node doesn't exist (while it would have been created by "replace"). This is not the same behaviour as with "tc" where "change" can be used to modify the value of some options while leaving the others unchanged. However, as not all entities support this operation, we believe the "change" commands as implemented here are more useful.

Also available "modules" (returns tc plugins dict) and "help" commands:

help(ip.tc("modules")["htb"])
print(ip.tc("help", "htb"))
route(command, **kwarg)

Route operations.

Keywords to set up rtmsg fields:

  • dst_len, src_len -- destination and source mask(see dst below)

  • tos -- type of service

  • table -- routing table

  • proto -- redirect, boot, static (see rt_proto)

  • scope -- routing realm

  • type -- unicast, local, etc. (see rt_type)

pyroute2/netlink/rtnl/rtmsg.py rtmsg.nla_map:

  • table -- routing table to use (default: 254)

  • gateway -- via address

  • prefsrc -- preferred source IP address

  • dst -- the same as prefix

  • iif -- incoming traffic interface

  • oif -- outgoing traffic interface

etc.

One can specify mask not as dst_len, but as a part of dst, e.g.: dst="10.0.0.0/24".

Commands:

add

Example:

ipr.route("add", dst="10.0.0.0/24", gateway="192.168.0.1")

...

More route() examples. Blackhole route:

ipr.route(
    "add",
    dst="10.0.0.0/24",
    type="blackhole",
)

Create a route with metrics:

ipr.route(
    "add",
    dst="172.16.0.0/24",
    gateway="10.0.0.10",
    metrics={
        "mtu": 1400,
        "hoplimit": 16,
    },
)

Multipath route:

ipr.route(
    "add",
    dst="10.0.0.0/24",
    multipath=[
        {"gateway": "192.168.0.1", "hops": 2},
        {"gateway": "192.168.0.2", "hops": 1},
        {"gateway": "192.168.0.3"},
    ],
)

MPLS lwtunnel on eth0:

ipr.route(
    "add",
    dst="10.0.0.0/24",
    oif=ip.link_lookup(ifname="eth0"),
    encap={
        "type": "mpls",
        "labels": "200/300",
    },
)

IPv6 next hop for IPv4 dst:

ipr.route(
    "add",
    prefsrc="10.127.30.4",
    dst="172.16.0.0/24",
    via={"family": AF_INET6, "addr": "fe80::1337"},
    oif=ipr.link_lookup(ifname="eth0"),
    table=100,
)

Create MPLS route: push label:

# $ sudo modprobe mpls_router
# $ sudo sysctl net.mpls.platform_labels=1024
ipr.route(
    "add",
    family=AF_MPLS,
    oif=ipr.link_lookup(ifname="eth0"),
    dst=0x200,
    newdst=[0x200, 0x300],
)

MPLS multipath:

ipr.route(
    "add",
    dst="10.0.0.0/24",
    table=20,
    multipath=[
        {
            "gateway": "192.168.0.1",
            "encap": {"type": "mpls", "labels": 200},
        },
        {
            "ifindex": ipr.link_lookup(ifname="eth0"),
            "encap": {"type": "mpls", "labels": 300},
        },
    ],
)

MPLS target can be int, string, dict or list:

"labels": 300    # simple label
"labels": "300"  # the same
"labels": (200, 300)  # stacked
"labels": "200/300"   # the same

# explicit label definition
"labels": {
    "bos": 1,
    "label": 300,
    "tc": 0,
    "ttl": 16,
}

Create SEG6 tunnel encap mode (kernel >= 4.10):

ipr.route(
    "add",
    dst="2001:0:0:10::2/128",
    oif=idx,
    encap={
        "type": "seg6",
        "mode": "encap",
        "segs": "2000::5,2000::6",
    },
)

Create SEG6 tunnel inline mode (kernel >= 4.10):

ipr.route(
    "add",
    dst="2001:0:0:10::2/128",
    oif=idx,
    encap={
        "type": "seg6",
        "mode": "inline",
        "segs": ["2000::5", "2000::6"],
    },
)

Create SEG6 tunnel inline mode with hmac (kernel >= 4.10):

ipr.route(
    "add",
    dst="2001:0:0:22::2/128",
    oif=idx,
    encap={
        "type": "seg6",
        "mode": "inline",
        "segs": "2000::5,2000::6,2000::7,2000::8",
        "hmac": 0xf,
    },
)

Create SEG6 tunnel with ip4ip6 encapsulation (kernel >= 4.14):

ipr.route(
    "add",
    dst="172.16.0.0/24",
    oif=idx,
    encap={
        "type": "seg6",
        "mode": "encap",
        "segs": "2000::5,2000::6",
    },
)

Create SEG6LOCAL tunnel End.DX4 action (kernel >= 4.14):

ipr.route(
    "add",
    dst="2001:0:0:10::2/128",
    oif=idx,
    encap={
        "type": "seg6local",
        "action": "End.DX4",
        "nh4": "172.16.0.10",
    },
)

Create SEG6LOCAL tunnel End.DT6 action (kernel >= 4.14):

ipr.route(
    "add",
    dst="2001:0:0:10::2/128",
    oif=idx,
    encap={
        "type": "seg6local",
        "action": "End.DT6",
        "table": "10",
    },
)

Create SEG6LOCAL tunnel End.DT4 action (kernel >= 5.11):

# $ sudo modprobe vrf
# $ sudo sysctl -w net.vrf.strict_mode=1
ipr.link(
    "add",
    ifname="vrf-foo",
    kind="vrf",
    vrf_table=10,
)
ipr.route(
    "add",
    dst="2001:0:0:10::2/128",
    oif=idx,
    encap={
        "type": "seg6local",
        "action": "End.DT4",
        "vrf_table": 10,
    },
)

Create SEG6LOCAL tunnel End.B6 action (kernel >= 4.14):

ipr.route(
    "add",
    dst="2001:0:0:10::2/128",
    oif=idx,
    encap={
        "type": "seg6local",
        "action": "End.B6",
        "srh": {"segs": "2000::5,2000::6"},
    },
)

Create SEG6LOCAL tunnel End.B6 action with hmac (kernel >= 4.14):

ipr.route(
    "add",
    dst="2001:0:0:10::2/128",
    oif=idx,
    encap={
        "type": "seg6local",
        "action": "End.B6",
        "srh": {
            "segs": "2000::5,2000::6",
            "hmac": 0xf,
        },
    },
)

change, replace, append

Commands change, replace and append have the same meanings as in ip-route(8): change modifies only existing route, while replace creates a new one, if there is no such route yet. append allows to create an IPv6 multipath route.

del

Remove the route. The same syntax as for add.

get

Get route by spec.

dump

Dump all routes.

rule(command, **kwarg)

Rule operations

  • command — add, delete

  • table — 0 < table id < 253

  • priority — 0 < rule's priority < 32766

  • action — type of rule, default 'FR_ACT_NOP' (see fibmsg.py)

  • rtscope — routing scope, default RT_SCOPE_UNIVERSE

    (RT_SCOPE_UNIVERSE|RT_SCOPE_SITE| RT_SCOPE_LINK|RT_SCOPE_HOST|RT_SCOPE_NOWHERE)

  • family — rule's family (socket.AF_INET (default) or

    socket.AF_INET6)

  • src — IP source for Source Based (Policy Based) routing's rule

  • dst — IP for Destination Based (Policy Based) routing's rule

  • src_len — Mask for Source Based (Policy Based) routing's rule

  • dst_len — Mask for Destination Based (Policy Based) routing's

    rule

  • iifname — Input interface for Interface Based (Policy Based)

    routing's rule

  • oifname — Output interface for Interface Based (Policy Based)

    routing's rule

  • uid_range — Range of user identifiers, a string like "1000:1234"

  • dport_range — Range of destination ports, a string like "80-120"

  • sport_range — Range of source ports, as a string like "80-120"

All packets route via table 10:

# 32000: from all lookup 10
# ...
ip.rule('add', table=10, priority=32000)

Default action:

# 32001: from all lookup 11 unreachable
# ...
iproute.rule('add',
             table=11,
             priority=32001,
             action='FR_ACT_UNREACHABLE')

Use source address to choose a routing table:

# 32004: from 10.64.75.141 lookup 14
# ...
iproute.rule('add',
             table=14,
             priority=32004,
             src='10.64.75.141')

Use dst address to choose a routing table:

# 32005: from 10.64.75.141/24 lookup 15
# ...
iproute.rule('add',
             table=15,
             priority=32005,
             dst='10.64.75.141',
             dst_len=24)

Match fwmark:

# 32006: from 10.64.75.141 fwmark 0xa lookup 15
# ...
iproute.rule('add',
             table=15,
             priority=32006,
             dst='10.64.75.141',
             fwmark=10)
stats(command, **kwarg)

Stats prototype.

class pyroute2.iproute.linux.IPBatch(*argv, **kwarg)

Netlink requests compiler. Does not send any requests, but instead stores them in the internal binary buffer. The contents of the buffer can be used to send batch requests, to test custom netlink parsers and so on.

Uses RTNL_API and provides all the same API as normal IPRoute objects:

# create the batch compiler
ipb = IPBatch()
# compile requests into the internal buffer
ipb.link("add", index=550, ifname="test", kind="dummy")
ipb.link("set", index=550, state="up")
ipb.addr("add", index=550, address="10.0.0.2", mask=24)
# save the buffer
data = ipb.batch
# reset the buffer
ipb.reset()
...
# send the buffer
IPRoute().sendto(data, (0, 0))
class pyroute2.iproute.linux.IPRoute(*argv, **kwarg)

Regular ordinary utility class, see RTNL API for the list of methods.

class pyroute2.iproute.linux.RawIPRoute(*argv, **kwarg)

The same as IPRoute, but does not use the netlink proxy. Thus it can not manage e.g. tun/tap interfaces.

class pyroute2.iproute.linux.ChaoticIPRoute(*argv, **kwarg)

IPRoute interface for chaotic tests - raising exceptions randomly.

Queueing disciplines

drr

The qdisc doesn't accept any parameters, but the class accepts quantum parameter:

ip.tc('add', 'drr', interface, '1:')
ip.tc('add-class', 'drr', interface, '1:10', quantum=1600)
ip.tc('add-class', 'drr', interface, '1:20', quantum=1600)
class pyroute2.netlink.rtnl.tcmsg.sched_drr.options(data=None, offset=0, length=None, parent=None, init=None)
class pyroute2.netlink.rtnl.tcmsg.sched_drr.stats(data=None, offset=0, length=None, parent=None, init=None)
class pyroute2.netlink.rtnl.tcmsg.sched_drr.stats2(data=None, offset=0, length=None, parent=None, init=None)
class stats_app(data=None, offset=0, length=None, parent=None, init=None)

choke

Parameters:

  • limit (required) -- int

  • bandwith (required) -- str/int

  • min -- int

  • max -- int

  • avpkt -- str/int, packet size

  • burst -- int

  • probability -- float

  • ecn -- bool

Example:

ip.tc('add', 'choke', interface,
      limit=5500,
      bandwith="10mbit",
      ecn=True)
class pyroute2.netlink.rtnl.tcmsg.sched_choke.options(data=None, offset=0, length=None, parent=None, init=None)
class qopt(data=None, offset=0, length=None, parent=None, init=None)
class stab(data=None, offset=0, length=None, parent=None, init=None)
encode()

Encode the message into the binary buffer:

msg.encode()
sock.send(msg.data)

If you want to customize the encoding process, override the method:

class CustomMessage(nlmsg):

    def encode(self):
        ...  # do some custom data tuning
        nlmsg.encode(self)
class pyroute2.netlink.rtnl.tcmsg.sched_choke.stats(data=None, offset=0, length=None, parent=None, init=None)
class pyroute2.netlink.rtnl.tcmsg.sched_choke.stats2(data=None, offset=0, length=None, parent=None, init=None)
class stats_app(data=None, offset=0, length=None, parent=None, init=None)

clsact

The clsact qdisc provides a mechanism to attach integrated filter-action classifiers to an interface, either at ingress or egress, or both. The use case shown here is using a bpf program (implemented elsewhere) to direct the packet processing. The example also uses the direct-action feature to specify what to do with each packet (pass, drop, redirect, etc.).

BPF ingress/egress example using clsact qdisc:

# open_bpf_fd is outside the scope of pyroute2
#fd = open_bpf_fd()
eth0 = ip.get_links(ifname="eth0")[0]
ip.tc("add", "clsact", eth0)
# add ingress clsact
ip.tc("add-filter", "bpf", idx, ":1", fd=fd, name="myprog",
      parent="ffff:fff2", classid=1, direct_action=True)
# add egress clsact
ip.tc("add-filter", "bpf", idx, ":1", fd=fd, name="myprog",
      parent="ffff:fff3", classid=1, direct_action=True)

hfsc

Simple HFSC example:

eth0 = ip.get_links(ifname="eth0")[0]
ip.tc("add", "hfsc", eth0,
      handle="1:",
      default="1:1")
ip.tc("add-class", "hfsc", eth0,
      handle="1:1",
      parent="1:0"
      rsc={"m2": "5mbit"})

HFSC curve nla types:

  • rsc: real-time curve

  • fsc: link-share curve

  • usc: upper-limit curve

class pyroute2.netlink.rtnl.tcmsg.sched_hfsc.options_hfsc(data=None, offset=0, length=None, parent=None, init=None)
class pyroute2.netlink.rtnl.tcmsg.sched_hfsc.options_hfsc_class(data=None, offset=0, length=None, parent=None, init=None)
class hfsc_curve(data=None, offset=0, length=None, parent=None, init=None)
class pyroute2.netlink.rtnl.tcmsg.sched_hfsc.stats2(data=None, offset=0, length=None, parent=None, init=None)
class stats_app(data=None, offset=0, length=None, parent=None, init=None)

htb

TODO: list parameters

An example with htb qdisc, lets assume eth0 == 2:

#          u32 -->    +--> htb 1:10 --> sfq 10:0
#          |          |
#          |          |
# eth0 -- htb 1:0 -- htb 1:1
#          |          |
#          |          |
#          u32 -->    +--> htb 1:20 --> sfq 20:0

eth0 = 2
# add root queue 1:0
ip.tc("add", "htb", eth0, 0x10000, default=0x200000)

# root class 1:1
ip.tc("add-class", "htb", eth0, 0x10001,
      parent=0x10000,
      rate="256kbit",
      burst=1024 * 6)

# two branches: 1:10 and 1:20
ip.tc("add-class", "htb", eth0, 0x10010,
      parent=0x10001,
      rate="192kbit",
      burst=1024 * 6,
      prio=1)
ip.tc("add-class", "htb", eht0, 0x10020,
      parent=0x10001,
      rate="128kbit",
      burst=1024 * 6,
      prio=2)

# two leaves: 10:0 and 20:0
ip.tc("add", "sfq", eth0, 0x100000,
      parent=0x10010,
      perturb=10)
ip.tc("add", "sfq", eth0, 0x200000,
      parent=0x10020,
      perturb=10)

# two filters: one to load packets into 1:10 and the
# second to 1:20
ip.tc("add-filter", "u32", eth0,
      parent=0x10000,
      prio=10,
      protocol=socket.AF_INET,
      target=0x10010,
      keys=["0x0006/0x00ff+8", "0x0000/0xffc0+2"])
ip.tc("add-filter", "u32", eth0,
      parent=0x10000,
      prio=10,
      protocol=socket.AF_INET,
      target=0x10020,
      keys=["0x5/0xf+0", "0x10/0xff+33"])
class pyroute2.netlink.rtnl.tcmsg.sched_htb.stats(data=None, offset=0, length=None, parent=None, init=None)
class pyroute2.netlink.rtnl.tcmsg.sched_htb.qdisc_stats2(data=None, offset=0, length=None, parent=None, init=None)
class pyroute2.netlink.rtnl.tcmsg.sched_htb.class_stats2(data=None, offset=0, length=None, parent=None, init=None)
class stats_app(data=None, offset=0, length=None, parent=None, init=None)
class pyroute2.netlink.rtnl.tcmsg.sched_htb.options(data=None, offset=0, length=None, parent=None, init=None)
class htb_glob(data=None, offset=0, length=None, parent=None, init=None)
class htb_parms(data=None, offset=0, length=None, parent=None, init=None)

Filters

u32

Filters can take an action argument, which affects the packet behavior when the filter matches. Currently the gact, bpf, and police action types are supported, and can be attached to the u32 and bpf filter types:

# An action can be a simple string, which translates to a gact type
action = "drop"

# Or it can be an explicit type (these are equivalent)
action = dict(kind="gact", action="drop")

# There can also be a chain of actions, which depend on the return
# value of the previous action.
action = [
    dict(kind="bpf", fd=fd, name=name, action="ok"),
    dict(kind="police", rate="10kbit", burst=10240, limit=0),
    dict(kind="gact", action="ok"),
]

# Add the action to a u32 match-all filter
ip.tc("add", "htb", eth0, 0x10000, default=0x200000)
ip.tc("add-filter", "u32", eth0,
      parent=0x10000,
      prio=10,
      protocol=protocols.ETH_P_ALL,
      target=0x10020,
      keys=["0x0/0x0+0"],
      action=action)

# Add two more filters: One to send packets with a src address of
# 192.168.0.1/32 into 1:10 and the second to send packets  with a
# dst address of 192.168.0.0/24 into 1:20
ip.tc("add-filter", "u32", eth0,
    parent=0x10000,
    prio=10,
    protocol=protocols.ETH_P_IP,
    target=0x10010,
    keys=["0xc0a80001/0xffffffff+12"])
    # 0xc0a800010 = 192.168.0.1
    # 0xffffffff = 255.255.255.255 (/32)
    # 12 = Source network field bit offset

ip.tc("add-filter", "u32", eth0,
    parent=0x10000,
    prio=10,
    protocol=protocols.ETH_P_IP,
    target=0x10020,
    keys=["0xc0a80000/0xffffff00+16"])
    # 0xc0a80000 = 192.168.0.0
    # 0xffffff00 = 255.255.255.0 (/24)
    # 16 = Destination network field bit offset