Skip to main content

cidrsubnet Function

cidrsubnet calculates a subnet address within given IP network address prefix.

cidrsubnet(prefix, newbits, netnum)

prefix must be given in CIDR notation, as defined in RFC 4632 section 3.1.

newbits is the number of additional bits with which to extend the prefix. For example, if given a prefix ending in /16 and a newbits value of 4, the resulting subnet address will have length /20.

netnum is a whole number that can be represented as a binary integer with no more than newbits binary digits, which will be used to populate the additional bits added to the prefix.

This function accepts both IPv6 and IPv4 prefixes, and the result always uses the same addressing scheme as the given prefix.

Unlike the related function cidrsubnets, cidrsubnet allows you to give a specific network number to use. cidrsubnets can allocate multiple network addresses at once, but numbers them automatically starting with zero.

Examples

> cidrsubnet("172.16.0.0/12", 4, 2)
172.18.0.0/16
> cidrsubnet("10.1.2.0/24", 4, 15)
10.1.2.240/28
> cidrsubnet("fd00:fd12:3456:7890::/56", 16, 162)
fd00:fd12:3456:7800:a200::/72

Netmasks and Subnets

Using cidrsubnet requires familiarity with some network addressing concepts.

The most important idea is that an IP address (whether IPv4 or IPv6) is fundamentally constructed from binary digits, even though we conventionally represent it as either four decimal octets (for IPv4) or a sequence of 16-bit hexadecimal numbers (for IPv6).

Taking our example above of cidrsubnet("10.1.2.0/24", 4, 15), the function will first convert the given IP address string into an equivalent binary representation:

      10 .        1 .        2 .        0
00001010 00000001 00000010 | 00000000
network | host

The /24 at the end of the prefix string specifies that the first 24 bits -- or, the first three octets -- of the address identify the network while the remaining bits (32 - 24 = 8 bits in this case) identify hosts within the network.

The CLI tool ipcalc is useful for visualizing CIDR prefixes as binary numbers. We can confirm the conversion above by providing the same prefix string to ipcalc:

$ ipcalc 10.1.2.0/24
Address: 10.1.2.0 00001010.00000001.00000010. 00000000
Netmask: 255.255.255.0 = 24 11111111.11111111.11111111. 00000000
Wildcard: 0.0.0.255 00000000.00000000.00000000. 11111111
=>
Network: 10.1.2.0/24 00001010.00000001.00000010. 00000000
HostMin: 10.1.2.1 00001010.00000001.00000010. 00000001
HostMax: 10.1.2.254 00001010.00000001.00000010. 11111110
Broadcast: 10.1.2.255 00001010.00000001.00000010. 11111111
Hosts/Net: 254 Class A, Private Internet

This gives us some additional information but also confirms (using a slightly different notation) the conversion from decimal to binary and shows the range of possible host addresses in this network.

While cidrhost allows calculating single host IP addresses, cidrsubnet on the other hand creates a new network prefix within the given network prefix. In other words, it creates a subnet.

When we call cidrsubnet we also pass two additional arguments: newbits and netnum. newbits decides how much longer the resulting prefix will be in bits; in our example here we specified 4, which means that the resulting subnet will have a prefix length of 24 + 4 = 28 bits. We can imagine these bits breaking down as follows:

      10 .        1 .        2 .    ?        0
00001010 00000001 00000010 | XXXX | 0000
parent network | netnum | host

Four of the eight bits that were originally the "host number" are now being repurposed as the subnet number. The network prefix no longer falls on an exact octet boundary, so in effect we are now splitting the last decimal number in the IP address into two parts, using half of it to represent the subnet number and the other half to represent the host number.

The netnum argument then decides what number value to encode into those four new subnet bits. In our current example we passed 15, which is represented in binary as 1111, allowing us to fill in the XXXX segment in the above:

      10 .        1 .        2 .    15       0
00001010 00000001 00000010 | 1111 | 0000
parent network | netnum | host

To convert this back into normal decimal notation we need to recombine the two portions of the final octet. Converting 11110000 from binary to decimal gives 240, which can then be combined with our new prefix length of 28 to produce the result 10.1.2.240/28. Again we can pass this prefix string to ipcalc to visualize it:

$ ipcalc 10.1.2.240/28
Address: 10.1.2.240 00001010.00000001.00000010.1111 0000
Netmask: 255.255.255.240 = 28 11111111.11111111.11111111.1111 0000
Wildcard: 0.0.0.15 00000000.00000000.00000000.0000 1111
=>
Network: 10.1.2.240/28 00001010.00000001.00000010.1111 0000
HostMin: 10.1.2.241 00001010.00000001.00000010.1111 0001
HostMax: 10.1.2.254 00001010.00000001.00000010.1111 1110
Broadcast: 10.1.2.255 00001010.00000001.00000010.1111 1111
Hosts/Net: 14 Class A, Private Internet

The new subnet has four bits available for host numbering, which means that there are 14 host addresses available for assignment once we subtract the network's own address and the broadcast address. You can thus use cidrhost function to calculate those host addresses by providing it a value between 1 and 14:

> cidrhost("10.1.2.240/28", 1)
10.1.2.241
> cidrhost("10.1.2.240/28", 14)
10.1.2.254

For more information on CIDR notation and subnetting, see Classless Inter-domain Routing.

  • cidrhost calculates the IP address for a single host within a given network address prefix.
  • cidrnetmask converts an IPv4 network prefix in CIDR notation into netmask notation.
  • cidrsubnets can allocate multiple consecutive addresses under a prefix at once, numbering them automatically.