IPv4 Datagrams: Headers, TTL, MTU, and Fragmentation

The IPv4 Datagram Format

A network-layer packet is called a datagram. An IPv4 datagram consists of a header and data. The fixed part of the header is 20 bytes long and may be followed by variable-length options; the data is usually a TCP segment or a UDP datagram. The figure lays out the header 32 bits per row. Click a field to see what it does.

The protocol field plays a role similar to a transport-layer port number. Ports deliver data to the right process; the protocol field delivers a datagram's data to the right upper-layer protocol.

TTL and the Header Checksum

If routing ever forms a loop, a datagram with no safeguard would circle the network forever. The time-to-live (TTL) field prevents this. Each router decrements TTL by 1; when TTL reaches 0, the router discards the datagram and usually returns an ICMP time-exceeded message to the source (see Section 4.6).

The header checksum is computed the same way as the UDP checksum in Chapter 3: treat the header as a sequence of 16-bit words, wrap any overflow around while summing, and complement the result. Because TTL changes at every hop, every router must recompute the header checksum and write it back.

Here $w_i$ is the $i$-th 16-bit word of the header (with the checksum field set to 0 while computing), fold wraps any carry beyond 16 bits back into the low 16 bits, and the right-hand equation is the receiver's check.

IP checks only the header; errors in the data are detected by upper-layer protocols such as TCP and UDP. Routers therefore process only about 20 header bytes, which keeps forwarding fast.

MTU and Fragmentation

The maximum amount of data a link-layer frame can carry is the maximum transmission unit (MTU). Ethernet's MTU, for example, is 1500 bytes. The links along an end-to-end path may use different link-layer protocols with different MTUs. When a datagram's total length exceeds the MTU of the next link, the router divides it into several smaller datagrams called fragments.

Each fragment carries its own IP header, filled in according to these rules:

1. Every fragment copies the original identification field, so the receiver knows which fragments belong together.
2. Every fragment except the last carries a multiple of 8 data bytes.
3. The fragment offset is the starting byte of the fragment's data within the original data, divided by 8.
4. Every fragment except the last has MF = 1.

For example, a datagram with total length 4000 bytes (20-byte header, 3980 data bytes) must cross a link with MTU 1500 bytes. Each fragment can carry at most 1480 data bytes, so the datagram becomes three fragments:

Fragments may take different paths, so only the destination host reassembles them; routers do not. If any fragment is lost, the destination eventually discards all fragments it has received, and the whole datagram must be retransmitted by an upper layer. Fragmentation also adds header overhead and consumes router resources. Modern hosts therefore usually set DF and perform path MTU discovery: a router that finds a datagram too large but with DF = 1 drops it and returns an ICMP "fragmentation needed" message, and the host reduces its datagram size. IPv6 goes further: routers never fragment at all (see Section 4.7).