Internet Standard Subnetting Procedure
updated by
- rfc-6918 — Formally Deprecating Some ICMPv4 Message Types
updates
- rfc-792 — Internet Control Message Protocol
also
- std-5
Extracted elements (20)
Subnetting was introduced to avoid the explosion in Internet routing table size caused by assigning a distinct network number to each physical LAN cable, keeping internal topology details out of global routing while still supporting multi-LAN organizations.
The address-mask scheme was chosen over self-encoding or fixed-width field schemes because it allows each organization to choose the best allocation of local address bits between subnet and host numbers, while costing no more to implement than any other approach.
The transparent subnet approach (RFC-925) was rejected as a general solution because it does not work for all LAN technologies (particularly those without ARP or broadcast support) and requires bridges to maintain caches that grow with total host count.
Since there is only one possible address mask value for a given subnet, a host need not match Address Mask Replies to its specific Request; the Identifier and Sequence Number fields can be ignored and multiple gateway replies cause no conflict.
When appropriate design choices are made in subnet mask assignment, hosts that believe they are on a non-subnetted network can be used on a subnetted one without modification, enabling gradual transition as described in RFC-917.
A gateway receiving an Address Mask Request should return the message with the 32-bit address mask field set to the mask identifying the subnet and network for the subnet on which the request was received.
If a host cannot determine the address mask after a reasonable number of tries, it should use a mask identical to the Internet network number mask (no subnetting assumed), as this is the safest default and will not prevent transmissions that would otherwise succeed.
If the source address in an Address Mask Request is zero (the requesting host does not know its IP address), the reply must be broadcast rather than sent directly to the requester.
No host or gateway should send an Address Mask Reply based on a guessed value of the address mask; only authoritative mask values should be sent.
The all-zeros and all-ones values in the subnet field must not be assigned to actual (physical) subnets, preserving the special meaning of zero ('this network') and all-ones ('all hosts') consistent with IP addressing conventions.
The subnet bits in an Internet address need not be contiguous, but it is recommended that they be contiguous and located as the most significant bits of the local address.
The use of the anonymous (0.0.0.0) source address in Address Mask Requests must be kept to a minimum, as broadcast replies impose unnecessary load on all hosts on the subnet.
When a gateway comes up it should broadcast an Address Mask Reply; when a host receives such a message that disagrees with its current mask, it should update its mask to conform to the received value.
For multiply-connected hosts, the my_ip_addr and my_ip_mask quantities must be maintained on a per-interface basis, and the routing conditional must be evaluated for each interface.
Internet addresses in a subnetted network are interpreted as <network-number><subnet-number><host-number>, where the subnet-number field width is constant for a given network and specified by an address mask. A zero-width subnet field means the network is not subnetted.
The ICMP Address Mask Request (AM1) is intended to be broadcast by a booting host to discover the subnet mask. A gateway or host acting in lieu of a gateway responds with an Address Mask Reply (AM2) containing the 32-bit mask.
The ICMP protocol is extended with two new message types: Address Mask Request (AM1, type 17) and Address Mask Reply (AM2, type 18), used for hosts to discover the subnet address mask via broadcast.
To support subnets, a host must store a 32-bit address mask (my_ip_mask) with bits set for the IP network number field and the subnet number field. Routing decisions use bitwise AND of the destination address with this mask, compared to the masked local address.
Two ICMP message type numbers are assigned: AM1 = 17 (Address Mask Request) and AM2 = 18 (Address Mask Reply), as listed in Appendix IV.
The ICMP Address Mask message is 96 bits: 8-bit Type (AM1=17 or AM2=18), 8-bit Code (0), 16-bit Checksum, 16-bit Identifier, 16-bit Sequence Number, and 32-bit Address Mask field.