Replacing IP addresses and spaces

Question

I'm trying to transform ifconfig -a to give specific output but I am not sure whether sed is performing well. There is a possibility that the particular version of sed isn't performing as it should (due to different sed counterpart).

My ifconfig -a output (I only want to see the netmask):

    xennet0: flags=8863<UP,BROADCAST,NOTRAILERS,RUNNING,SIMPLEX,MULTICAST> mtu 1500
            capabilities=2800<TCP4CSUM_Tx,UDP4CSUM_Tx>
            enabled=0
            address: 0a:3d:c0:98:c6:73
            inet 172.31.11.166 netmask 0xfffff000 broadcast 172.31.15.255
            inet6 fe80::83d:c0ff:fe98:c673%xennet0 prefixlen 64 scopeid 0x1
    lo0: flags=8049<UP,LOOPBACK,RUNNING,MULTICAST> mtu 33184
            inet 127.0.0.1 netmask 0xff000000
            inet6 ::1 prefixlen 128
            inet6 fe80::1%lo0 prefixlen 64 scopeid 0x2

Needed output (transformed output):

 0xfffff000
 0xff000000

My failed attempt:

ifconfig -a | egrep "0xf." | sed 's/inet //' | sed 's/netmask //' |
sed 's/ broadcast//' | sed 's/([0-9]{1,3}\.){3}[0-9]{1,3}//g' | sed 's/\s+//'

It gave me output as:

        172.31.11.166 0xfffff000 172.31.15.255
        127.0.0.1 0xff000000

I expected it to give my needed output.

Help will be of great assistance. I am using NetBSD 6.1.5 (Amazon EC2), however, I believe any general fixture should work.

Answer 1

With GNU grep, I'd use Perl regex with the -o option (retain only matches):

$ ifconfig -a | grep -Po '(?<=\bnetmask )\w*'
0xfffff000
0xff000000

For BSD grep, where we don't have the -P option, we can use awk or cut to get the second part only, as in

$ ifconfig -a | grep -o '\bnetmask [^ ]*' | cut -d ' ' -f 2
0xfffff000
0xff000000

If you really want to use sed, you could do this:

$ ifconfig -a | sed -n 's/.*\bnetmask \([^ ]*\).*/\1/p'
0xfffff000
0xff000000

This quite similar to Eric's answer. It suppresses output with -n, and for lines containing netmask, it extracts the word following it and prints it with (the p flag).

Answer 2

While doing it with sed might be interesting, I'd recommend awk as the right tool for this (and any similar) job:

ifconfig -a | awk '{ for (i = 1; i < NF; i++) {if ($i == "netmask") { print $(i+1); next; } }}'

Answer 3

In some respects, you're making it too hard. And using more commands than strictly required. Here's a command line that will do what you want with only one sed(1) invocation:

ifconfig -a | sed -ne '/inet /s/^.* netmask \(.*\) broadcast.*$/\1/p'

Explanation: the "/inet /" tells sed(1) to select only lines containing that string, and then apply the rest of the expression. The rest of the expression strips out the netmask from between the keywords "netmask " and " broadcast", and prints the result.

The "-n" on the sed(1) command line tells sed(1) not to print the lines as they're examined. The "-e" says the next argument is an inline sed(1) script.

Per my comment, that misses lo*.

Changing the pattern to:

'/inet /s/^.* netmask \(0x[0-9a-fA-F]*\).*$/\1/p'

creates the desired results. (including all interfaces running inet, including lo*)

Answer 4

Use Grep with Extended Regular Expressions

You can use grep's ERE engine and the --only-matching flag to extract just the hex values. To do this, use egrep or grep -E to turn on extended regular expressions. For example:

$ egrep -o '0x[[:xdigit:]]{8}' /tmp/corpus 
0xfffff000
0xff000000

Likewise, you can also pipe in output and get the same results with:

$ ifconfig -a | egrep -o '0x[[:xdigit:]]{8}'

This works with both GNU and BSD greps, and doesn't rely on the PCRE library. It should therefore work on most modern systems without the need to tweak anything.