The short answer

Here’s the invocation that works. It’s explained in the long answer that follows.

sort -n -t . -k 1,1 -k 2,2 -k 3,3 -k 4,4

A better short answer

A sharp-eyed reader notes that the sort binary included with GNU coreutils version 7.0 and higher accepts a new option: –version-sort (-V, –sort=version). Since IPv4 addresses look like software version numbers, the new option will reliably sort them:

sort -V

Since not all field-based sorting is that simple, I’ll continue with my original explanation – but for sorting IPv4 addresses the -V option is simplest.

The long answer

More than once I’ve been confronted with a list of IP addresses that I’ve wanted to sort into numeric order. Trouble is, the dotted-quad notation isn’t sort-friendly. Consider the following raw list of addresses.

$ cat addresses.txt
129.95.30.40
5.24.69.2
19.20.203.5
1.2.3.4
19.20.21.22
5.220.100.50

Without options, sort will rely on alphabetic order, which certainly won’t do what you want:

$ sort addresses.txt
1.2.3.4
129.95.30.40
19.20.203.5
19.20.21.22
5.220.100.50
5.24.69.2

There are so many mistakes in this ordering I’m not even going to try to list them all.

The situation is only marginally improved when using the --numeric-sort (-n) option.

$ sort -n addresses.txt
1.2.3.4
5.220.100.50
5.24.69.2
19.20.203.5
19.20.21.22
129.95.30.40

The first set of numbers in each dotted-quad sort correctly—5 preceeds 19, and 129 is at the tail end—but the internal numbering still gets improper treatment. 5.220.100.50 is listed prior to 5.24.69.2 because 220 is alphabetically prior to 24. Likewise the two 19.20.x.x addresses are mixed up because 203 is alphabetically prior to 21.

The solution is to tell sort to order the list numerically, considering each address as a set of four numeric fields, each separated by a dot.

$ sort -n -t . -k 1,1 -k 2,2 -k 3,3 -k 4,4 addresses.txt
1.2.3.4
5.24.69.2
5.220.100.50
19.20.21.22
19.20.203.5
129.95.30.40

In English, you’re saying, Yo, sort! I’ve got here a list of numbers (-n), but each item in the list consists of some subnumbers, fields set apart from the others by a dot (-t .). Sort first by the first field, and only the first field (-k 1,1), then by the second and only the second (-k 2,2), and so on (-k 3,3 -k 4,4).

Or, as I mentioned above, just use sort -V.

Similar operations

There are other widely used data with mixed numeric and alphabetic fields can be sorted with similar techniques.

The getent utility will sort group information by GID, but the sorting is done per-source. If you have an extensive /etc/group file and a large network-provided group database (from, e.g., LDAP), the groups are not interleaved. Here sort can do its magic:

getent group | sort -n -t: -k 3,3

When returning the passwd database, getent sorts by username not UID, making sort even more useful:

getent passwd | sort -n -t: -k 3,3

Howto