An IPv4 address is a 32-bit number, which is expressed as four 8-bit octets in a decimal format that ranges in value from 0 to 255, separated by single dots. The address is divided into two parts: the network prefix and the host number. The network prefix identifies a unique physical or virtual subnet. The host number identifies a specific host on the subnet. All hosts on the same subnet have the same network prefix and can talk to each other directly without a router. A network gateway connects different networks, and a network router commonly operates as the gateway for a subnet.

In the original IPv4 specification, the allowed network prefixes were one of three fixed sizes for unicast packets that have a single source and destination. The network prefix might be 8 bits (class A), 16 bits (class B), or 24 bits (class C). Today, the number of bits in the network prefix is variable, which means that the prefix can be any number in the supported range, and this later specification is called Classless Inter-Domain Routing (CIDR). Although fixed-address classes are no longer in use, many network professionals still refer to networks with 8-bit, 16-bit, or 24-bit network prefixes by using the original class A, B, or C designation.

A network mask (netmask) is a binary mask whose length indicates how many bits belong to the network prefix that identifies the subnet. Because an IPv4 address is always 32 bits long, a subnet with a longer network mask has fewer available bits to identify hosts, which means fewer possible hosts. A subnet with a shorter network mask has more available bits to identify hosts, which means more possible hosts and a larger subnet.

Network masks are expressed in one of two forms, which are both used routinely. The first form, which is known as CIDR notation, appends a forward slash (/) and an integer up to 32 that indicates the number of bits in the binary mask. The second notation displays the number of bits in the binary mask as four 8-bit octets in decimal format.

The number of available host addresses in a subnet depends on the network prefix size. For example, a network prefix of /24 leaves 8 bits, or 255 possible host addresses in the subnet. A network prefix of /16 leaves 16 bits, or 65536 possible host addresses in the subnet.

The following figures illustrate the use of an IP address and a netmask to calculate the network prefix and the host number for a subnet. Perform a binary AND calculation where each bit in the IP address and netmask is binary, and compare each bit to its corresponding bit in the IP address and netmask through the prefix length. In an AND calculation, both bits must be a '1' for the result to be a '1', and all other combinations result in '0'. Perform a binary OR calculation on the remaining bits in the host number, where either bit can be a '1' for the result to be a '1'. In a binary OR calculation, only two '0' bits result in a '0'.

Figure 11.2: IPv4 netmask calculation for a small network

Figure 11.3: IPv4 netmask calculation for a larger network

In the following example, identify the netmask first, and then perform the binary calculations. A netmask of /24 means that the leading 24 bits of the address define the network address (192.168.1.0). In this scenario, 8 bits, or 254 addresses, are available for host addressing.

In the following example, a /19 netmask is a valid network prefix that uses only a partial octet. Variable length netmasks allow subnets with a different quantity of hosts than the full-octet netmasks. The remaining 13 bits, or 8190 addresses, are available for host addressing.

In the following example, the /8 indicates a large network. Only the first octet is used for the network prefix (10.0.0.0). The remaining 24 bits, or 16,777,214 addresses, are available for host addressing. The 10.255.255.255 broadcast address is the last address of the network.

Network packets move from host to host on a subnet and through routers from network to network. Each host has a routing table, which determines which network interface is correct for sending packets to particular networks. A routing table entry lists the destination network, which network interface to use, and the IP address of the router to forward the packet to the final destination. The routing table entry that matches the network prefix of the destination address is used to route the packet. If multiple entries are valid for the destination address, then the entry with the longer prefix is used.

If the destination network does not match a more specific entry, then the packet is routed by using the 0.0.0.0/0 default entry. This default route points to the gateway router on a local subnet that the host can reach.

When a router receives packets that are not addressed to itself, the router forwards the traffic based on its own routing table. Forwarding might send the packet directly to the destination host if this router is on the destination's subnet, or might forward the packet again to another router network. Packet forwarding on routers continues until the packet reaches the requested destination network and host.

Figure 11.4: Example network topology

Consider the preceding network diagram and network routing table.

A server can automatically configure its IPv4 network settings by communicating with a DHCP server. A local DHCP client queries the subnet by using a link layer protocol to locate a DHCP server or proxy, and negotiates to use a unique address and other settings for a specific lease period. The client must periodically request lease renewal to maintain use of the assigned network configuration.

As an alternative, you can configure a server to use a static network configuration. Static network settings are read from local configuration files. The settings that you use must be appropriate for your subnet. Coordinate with your network administrator to avoid conflicts with other servers in the same subnets.

An IPv6 address is a 128-bit number, which is normally expressed as eight colon-separated groups of four hexadecimal nibbles (half-bytes). Each nibble represents four bits of the IPv6 address, so each group represents 16 bits of the IPv6 address.

2001:0db8:0000:0010:0000:0000:0000:0001

To simplify writing IPv6 addresses, leading zeros in a colon-separated group are not needed. However, at least one hexadecimal digit must be written in each colon-separated group.

2001:db8:0:10:0:0:0:1

Because addresses with long strings of zeros are common, one or more consecutive groups of zeros only can be combined with exactly one block of two colon :: characters.

2001:db8:0:10::1

The 2001:db8::0010:0:0:0:1 IPv6 address, though a valid representation, is a less convenient way to write the example address. This different representation can confuse administrators who are new to IPv6. The following list shows tips for writing consistently readable addresses:

[2001:db8:0:10::1]:80

A normal IPv6 unicast address is divided into two parts: the network prefix and interface ID. The network prefix identifies the subnet. Two network interfaces on the same subnet cannot have the same interface ID; the interface ID identifies a particular interface on the subnet.

Unlike IPv4, IPv6 has a standard subnet mask, /64, which is used for almost all normal addresses. In this case, half of the 128 bit address is the network prefix and the other half is the interface ID. With 64 bits for host addresses, a single subnet could theoretically contain 2^64 hosts.

Typically, the network provider allocates a shorter prefix to an organization, such as a /48. This prefix leaves the rest of the network part for assigning subnets (up to the maximum /64 length) from that allocated prefix. For example, when a /48 allocation prefix is assigned, 16 bits are available for up to 65536 subnets.

Figure 11.5: IPv6 address parts and subnetting

A link-local address in IPv6 is an unroutable address that the system uses only to talk to other systems on the same network link. To ensure that the IP address is unique, the system uses a specific method to compute the interface ID of the link-local address.

Use the ip addr show command to retrieve the link-local IPv6 address, as in the following example. Add the -br option for a brief listing of addresses only.

[user@host ~]$ ip addr show dev eth0
3: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc fq_codel state UP group default qlen 1000
    link/ether 52:54:00:01:fa:0a brd ff:ff:ff:ff:ff:ff
    inet 10.42.0.1/16 brd 10.42.255.255 scope global noprefixroute eth0
       valid_lft forever preferred_lft forever
    inet6 fe80::7418:cf98:c742:3681/64 scope link noprefixroute
       valid_lft forever preferred_lft forever

Add the -br option for a brief listing of IPv4 and IPv6 addresses only.

[user@host ~]$ ip -br addr show dev eth0
eth0          UP          10.42.0.1/16 fe80::7418:cf98:c742:3681/64

To operate correctly, IPv6 relies on the link-local address. The interface always keeps that address, even when you assign a routable IPv6 address manually or with an automated method.

With multicast, one system can send traffic to a special IP address that multiple systems receive. Multicast differs from broadcast, because broadcast packets are not routable and reach only local subnet hosts. Conversely, multicast packets are routed to specific hosts that announced a request for the uniquely addressed multicast packets. Multicast packets can be routed to other subnets, if all intermediary routers support forwarding multicast requests and routing.

Multicast plays a larger role in IPv6 than in IPv4, because IPv6 has no broadcast address. The ff02::1 IPv6 address is a key multicast address that is used as the all-nodes link-local address, and behaves like a broadcast address. You can ping this address to send traffic to all nodes on the link. Link-scope multicast addresses (which start with ff02::/8) must be specified with a scope identifier, as for a link-local address.

[user@host ~]$ ping6 ff02::1%ens3
PING ff02::1%ens3(ff02::1) 56 data bytes
64 bytes from fe80::211:22ff:feaa:bbcc: icmp_seq=1 ttl=64 time=0.072 ms
64 bytes from fe80::200:aaff:fe33:2211: icmp_seq=1 ttl=64 time=102 ms (DUP!)
64 bytes from fe80::bcd:efff:fea1:b2c3: icmp_seq=1 ttl=64 time=103 ms (DUP!)
64 bytes from fe80::211:22ff:feaa:bbcc: icmp_seq=2 ttl=64 time=0.079 ms
...output omitted...

IPv4 has two ways to configure addresses on network interfaces. The administrator can manually configure network addresses on interfaces, or addresses can be configured dynamically with DHCP. IPv6 supports manual configuration, and two methods of dynamic configuration, one of which is DHCPv6.

You can select interface IDs for static IPv6 addresses, similar to IPv4. In IPv4, two addresses on a network cannot be used: the lowest address, which is the network address; and the highest address, which is the broadcast address. In IPv6, two interface IDs are reserved, and cannot be used as normal host interface addresses:

DHCPv6 lease negotiations work differently from IPv4 DHCP, because DHCPv6 has no broadcast address. A host sends a DHCPv6 request from a link-local address to port 547/UDP on the dedicated ff02::1:2 all-dhcp-servers link-local multicast group. A listening DHCPv6 server can choose to reply with appropriate information to port 546/UDP on the client's provided link-local address.

The dhcp package in Red Hat Enterprise Linux 9 provides support for a DHCPv6 server.

In addition to DHCPv6, IPv6 also supports another dynamic configuration method, which is called Stateless Address Autoconfiguration (SLAAC). To use SLAAC, a host configures its interface with a link-local fe80::/64 address, and sends a "router solicitation" to the ff02::2 all-routers link-local multicast group. An IPv6 router on the local link responds to the host's link-local address with the subnet's previously configured network prefix and other relevant information. The host uses the provided network prefix with an interface ID that is constructed the same as for link-local addresses. The router periodically sends multicast updates (router advertisements) to confirm or to update the provided network information.

With the radvd package in Red Hat Enterprise Linux 9, a Red Hat Enterprise Linux based IPv6 router can provide SLAAC through router advertisements.