Linux/macOS | |
Windows |
Warning
netifaces needs a new maintainer. al45tair is no longer able to maintain it or make new releases due to work commitments.
It's been annoying me for some time that there's no easy way to get the address(es) of the machine's network interfaces from Python. There is a good reason for this difficulty, which is that it is virtually impossible to do so in a portable manner. However, it seems to me that there should be a package you can easy_install that will take care of working out the details of doing so on the machine you're using, then you can get on with writing Python code without concerning yourself with the nitty gritty of system-dependent low-level networking APIs.
This package attempts to solve that problem.
First you need to install it, which you can do by typing:
tar xvzf netifaces-0.10.8.tar.gz cd netifaces-0.10.8 python setup.py install
Note that you will need the relevant developer tools for your platform, as netifaces is written in C and installing this way will compile the extension.
Once that's done, you'll need to start Python and do something like the following:
>>> import netifaces
Then if you enter
>>> netifaces.interfaces()
['lo0', 'gif0', 'stf0', 'en0', 'en1', 'fw0']
you'll see the list of interface identifiers for your machine.
You can ask for the addresses of a particular interface by doing
>>> netifaces.ifaddresses('lo0')
{18: [{'addr': ''}], 2: [{'peer': '127.0.0.1', 'netmask': '255.0.0.0', 'addr': '127.0.0.1'}], 30: [{'peer': '::1', 'netmask': 'ffff:ffff:ffff:ffff:ffff:ffff:ffff:ffff', 'addr': '::1'}, {'peer': '', 'netmask': 'ffff:ffff:ffff:ffff::', 'addr': 'fe80::1%lo0'}]}
Hmmmm. That result looks a bit cryptic; let's break it apart and explain what each piece means. It returned a dictionary, so let's look there first:
{ 18: [...], 2: [...], 30: [...] }
Each of the numbers refers to a particular address family. In this case, we
have three address families listed; on my system, 18 is AF_LINK
(which means
the link layer interface, e.g. Ethernet), 2 is AF_INET
(normal Internet
addresses), and 30 is AF_INET6
(IPv6).
But wait! Don't use these numbers in your code. The numeric values here are system dependent; fortunately, I thought of that when writing netifaces, so the module declares a range of values that you might need. e.g.
>>> netifaces.AF_LINK
18
Again, on your system, the number may be different.
So, what we've established is that the dictionary that's returned has one
entry for each address family for which this interface has an address. Let's
take a look at the AF_INET
addresses now:
>>> addrs = netifaces.ifaddresses('lo0')
>>> addrs[netifaces.AF_INET]
[{'peer': '127.0.0.1', 'netmask': '255.0.0.0', 'addr': '127.0.0.1'}]
You might be wondering why this value is a list. The reason is that it's possible for an interface to have more than one address, even within the same family. I'll say that again: you can have more than one address of the same type associated with each interface.
Asking for "the" address of a particular interface doesn't make sense.
Right, so, we can see that this particular interface only has one address, and, because it's a loopback interface, it's point-to-point and therefore has a peer address rather than a broadcast address.
Let's look at a more interesting interface.
>>> addrs = netifaces.ifaddresses('en0')
>>> addrs[netifaces.AF_INET]
[{'broadcast': '10.15.255.255', 'netmask': '255.240.0.0', 'addr': '10.0.1.4'}, {'broadcast': '192.168.0.255', 'addr': '192.168.0.47'}]
This interface has two addresses (see, I told you...) Both of them are regular IPv4 addresses, although in one case the netmask has been changed from its default. The netmask may not appear on your system if it's set to the default for the address range.
Because this interface isn't point-to-point, it also has broadcast addresses.
Now, say we want, instead of the IP addresses, to get the MAC address; that is, the hardware address of the Ethernet adapter running this interface. We can do
>>> addrs[netifaces.AF_LINK]
[{'addr': '00:12:34:56:78:9a'}]
Note that this may not be available on platforms without getifaddrs(), unless
they happen to implement SIOCGIFHWADDR
. Note also that you just get the
address; it's unlikely that you'll see anything else with an AF_LINK
address.
Oh, and don't assume that all AF_LINK
addresses are Ethernet; you might, for
instance, be on a Mac, in which case:
>>> addrs = netifaces.ifaddresses('fw0')
>>> addrs[netifaces.AF_LINK]
[{'addr': '00:12:34:56:78:9a:bc:de'}]
No, that isn't an exceptionally long Ethernet MAC address---it's a FireWire address.
As of version 0.10.0, you can also obtain a list of gateways on your machine:
>>> netifaces.gateways()
{2: [('10.0.1.1', 'en0', True), ('10.2.1.1', 'en1', False)], 30: [('fe80::1', 'en0', True)], 'default': { 2: ('10.0.1.1', 'en0'), 30: ('fe80::1', 'en0') }}
This dictionary is keyed on address family---in this case, AF_INET
---and
each entry is a list of gateways as (address, interface, is_default)
tuples.
Notice that here we have two separate gateways for IPv4 (AF_INET
); some
operating systems support configurations like this and can either route packets
based on their source, or based on administratively configured routing tables.
For convenience, we also allow you to index the dictionary with the special
value 'default'
, which returns a dictionary mapping address families to the
default gateway in each case. Thus you can get the default IPv4 gateway with
>>> gws = netifaces.gateways()
>>> gws['default'][netifaces.AF_INET]
('10.0.1.1', 'en0')
Do note that there may be no default gateway for any given address family;
this is currently very common for IPv6 and much less common for IPv4 but it
can happen even for AF_INET
.
BTW, if you're trying to configure your machine to have multiple gateways for the same address family, it's a very good idea to check the documentation for your operating system very carefully, as some systems become extremely confused or route packets in a non-obvious manner.
I'm very interested in hearing from anyone (on any platform) for whom the
gateways()
method doesn't produce the expected results. It's quite
complicated extracting this information from the operating system (whichever
operating system we're talking about), and so I expect there's at least one
system out there where this just won't work.
It gets regular testing on OS X, Linux and Windows. It has also been used successfully on Solaris, and it's expected to work properly on other UNIX-like systems as well. If you are running something that is not supported, and wish to contribute a patch, please use Github to send a pull request.
It's an MIT-style license. See LICENSE.
Because someone released a fork of netifaces with the version 0.9.0. Hopefully skipping the version number should remove any confusion. In addition starting with 0.10.0 Python 3 is now supported and other features/bugfixes have been included as well. See the CHANGELOG for a more complete list of changes.