Droms R. The DHCP handbook

Подождите немного. Документ загружается.

An IP node sending an IP packet on an ethernet network uses a combination of

static and dynamic translation. If the IP destination address is an IP broadcast

address on the link to which the node is attached, it translates this to the ethernet

all-stations address by using a static translation. Otherwise, it uses Address

Resolution Protocol (ARP) to dynamically determine the link-layer address of the

recipient.

With ARP, a node sends a datagram to the ethernet all-stations address that makes

the request “please tell me the link-layer address that corresponds to this IP address.”

Every node on the link receives this request. If a node on the link has the IP address

mentioned in the request, it replies with its link-layer address. The sender uses this

link-layer address to send IP datagrams to the destination. The sender records the

destination hardware address in a local cache for future use. The entries in this cache

must be deleted periodically, to ensure that the hardware addresses are updated. The

lifetime of an ARP cache entry can be specified through DHCP.

The Internet Layer

The internet layer of the TCP/IP protocol suite is responsible for end-to-end delivery

of protocol messages between computers. Internet Protocol (IP) is the protocol from

the TCP/IP suite that implements the internet layer. Logically, an internet can be

thought of as a collection of independent network segments or physical networks,

interconnected by routers that are attached to two or more network segments. Hosts

connected to the network segments communicate by exchanging network messages

through the interconnecting routers.

The IP datagram is the basic message unit for data delivered by IP; a host forms an IP

datagram and identifies the destination to which the datagram should be delivered,

and the IP software on the host and routers cooperates to deliver the datagram to the

destination.

Network (IP) Addresses

Each computer that uses IP is assigned an IP address, which performs two functions:

• It uniquely identifies the computer.

• It specifies the network segment to which the computer is connected.

In contrast, link-layer addresses provide unique identification only and give no infor-

mation about location.

As Figure 4.2 shows, an IP address includes a network number and a host identifier. As a

special case, each network segment in an internet has its own network address, with

a unique network number and a host identifier of zero. The IP address for a

CHAPTER 4 Configuring TCP/IP Stacks46

006 3273 CH04 10/3/02 5:01 PM Page 46

computer is composed of the network number from the network segment to which

it is attached and a host part that is unique among all the computers on that

segment.

The Internet Layer 47

Network number Host identifier

FIGURE 4.2 The structure of an IP address.

Composing the IP address from a network number and a host identifier produces

some interesting features:

• It means IP datagrams can be forwarded to their destination by examining

only the network number and not the entire IP address. The amount of infor-

mation needed to forward datagrams depends on the number of network

segments in the internet—not on the number of attached computers.

•Acomputer’s IP address depends on the network segment to which it is

attached. Thus, when a computer is first attached to an internet, it must be

given an address from the network segment to which it is attached. If that

computer is then moved to a different network segment, it must be given a

new IP address.

• The host identifier of a computer’s IP address must be unique among all the

computers on its network segment. If two computers from the same network

use the same host identifier, they both use the same IP address, and their

network connections will be unreliable or unusable.

The related problems of correctly configuring a computer with an address that

depends on its location within the network and avoiding simultaneous use of the

same address by different computers were the initial motivation for DHCP. Prior to

the development of DHCP, a network administrator generally had to walk around

with a slip of paper or a spreadsheet listing available addresses. To add a new

computer to the network without DHCP, he or she had to consult the list of avail-

able addresses, select an unused address, carefully mark the address as “in use,” and

manually configure the computer with that address.

Not only is this procedure time-consuming and error-prone, it doesn’t scale well.

Imagine the potential for conflict in a large organization where multiple network

administrators try to assign IP addresses simultaneously from the same spreadsheet

or sheet of paper.

006 3273 CH04 10/3/02 5:01 PM Page 47

Subnetting

The original IP specification (RFC 790) defines three major address classes, which

identify the ways in which an IP address is split into a network number and a host

identifier. The format of each address class determines the number of networks avail-

able in that class and the number of hosts that can be attached to each network.

RFC 988 defines an additional address class, D, which is used for multicast

(described later in this chapter). The format of the four address classes and the

number of networks and hosts for each class are shown in Figure 4.3.

CHAPTER 4 Configuring TCP/IP Stacks48

1110D(N/A) (N/A)Multicast Address

110 Net partC2

(

2 million) 2

–2 (254)Host part

10 Net partB2

(16382) 2

–2 (65534)Host

0NetA2

–1 (127) 2

–2 (

16 million)Host part

Address

class Number of networks Number of hostsAddress format

FIGURE 4.3 The structure of Class A, B, C, and D addresses.

The IP address classes aren’t a good match with some network architectures. For

example, if a network segment with 400 computers uses a Class B address that can

accommodate more than 65,000 computers, the segment wastes more than 99% of

the available addresses.

Subnetting (as described in RFC 950, RFC 1878, RFC 1519) is a technique for dividing

Class A, B, or C addresses into smaller groups of IP addresses that more closely

match the addressing requirements for network segments. In a subnetted network,

the part of an IP address to the right of the network number is split into a subnet

address and the host identifier. The subnet address locates the network segment

within the collection of segments that share the same network number, and the host

identifier identifies the specific host on that network segment. Figure 4.4 illustrates

the format of a subnetted IP address.

In the original IP addressing scheme, the division between the network address and

the host identifier is defined by the first three bits of an IP address. How is the

format of a subnetted IP address defined? Each subnetted address must have an asso-

ciated subnet mask, which identifies which bits of the address make up the subnet

address and which are used as the host address. A subnet mask is a 32-bit number

with a 1 in every bit from the IP address that is to be used as the network number or

006 3273 CH04 10/3/02 5:01 PM Page 48

the subnet address and a 0 in every bit from the IP address that is to be used as the

host identifier.

The Internet Layer 49

Network number

Subnet

number

Host

identifier

FIGURE 4.4 The structure of a subnetted IP address.

NOTE

The subnet bits in a subnet mask are contiguous and include both the network number field

and the subnet field. The notation /24 indicates that the most significant 24 bits of the subnet

mask are set to 1.

The subnet mask is applied to IP addresses to determine whether two IP addresses are

part of the same subnet or different subnets. The subnet address can be extracted

from an IP address by computing the bitwise logical AND of the address and the

subnet mask. Two IP addresses are part of the same subnet if the resulting subnet

addresses are equal.

As discussed in the next section, computers that use IP as part of the IP datagram

delivery process use this equality test. The subnet mask is, therefore, just as impor-

tant to the correct operation of IP as the IP address; if a computer has an incorrect

subnet mask, it might be unable to correctly process some IP datagrams.

The use of a subnet mask has been extended in classless internet domain routing

(CIDR) addressing, as described in RFC 1519. CIDR addressing allows the use of a

subnet mask that is shorter than the network number field in the associated address.

A CIDR address identifies a block of network addresses; for example, the CIDR

address 192.168.4.0/22 refers to the four class C network addresses 192.168.4.0

through 192.168.7.0.

Datagram Delivery

Hosts and routers cooperate to deliver datagrams from the sending host to the desti-

nation host. The original sender of a datagram and each router along the datagram’s

path to its destination examine the datagram to determine its destination and then

forward the datagram to the next router on the way to the destination. The sender

must make an initial decision about delivery: Is the destination of this datagram on

the same network segment? If so, the source can deliver the datagram directly to the

destination. Otherwise, the source must deliver the datagram to a router for forward-

ing to the destination.

006 3273 CH04 10/3/02 5:01 PM Page 49

The sender determines whether the destination is on the same network segment by

applying its subnet mask to both its own address and the destination address. If the

resulting IP addresses match, the destination is on the same network segment.

Otherwise, the source must forward the datagram to a router. Thus, a computer that

uses TCP/IP must be configured with the address of at least one router in order to

deliver datagrams to destinations on different network segments.

A computer uses a routing table that contains the addresses of routers to be used for

specific destinations. Often, a computer has one or more entries in its routing table

and uses these entries if no specific router for a destination exists. These entries,

called default routes, identify default routers and must be inserted into the routing

table manually or automatically through DHCP or by another mechanism such as

router discovery (RFC 1256).

Multiple IP Networks on a Network Segment

Although the discussion of IP addressing and datagram delivery in the previous

sections describes a network segment as having a single IP network number, it is

possible to assign multiple IP networks to a single physical network segment. These

IP networks, sometimes referred to as shared or overlay segments, function as though

they are assigned to independent network segments. Datagrams that are sent from a

computer on one IP network to a computer on the other network on the same

segment must be forwarded through a router, even though both computers are on

the same physical network. Routers attached to the network segment have separate

routing table entries for each IP network, each of which points to the same network

interface.

Assigning multiple IP networks to a single network segment complicates the manage-

ment of DHCP servers because the server must know which IP networks are associ-

ated with a common network segment. In addition, the network architect may have

rules about which IP network on a network segment a computer should use, and the

server must follow those rules.

Multicast

IP includes a form of datagram delivery called multicast, in which a datagram is

delivered to more than one destination. Multicast differs from broadcast in that a

multicast datagram may be delivered to a subset of the computers on an internet,

rather than to all the computers on a single IP network.

Multicast is used in applications in which many hosts want to receive copies of data-

grams sent by a source without the overhead of sending a separate copy of the data

to each destination. Examples of applications that use multicast include digitized

audio/video conferencing, other collaborative applications, and routing protocols.

CHAPTER 4 Configuring TCP/IP Stacks50

006 3273 CH04 10/3/02 5:01 PM Page 50

Other Internet-Layer Parameters

Several other IP functions have operational parameters that must be correctly

configured.

TTL

Each datagram has a time to live (TTL) field that is used to detect situations such as a

datagram caught in a routing loop. The TTL may be explicitly specified to the IP soft-

ware. If it is not, the IP software uses a default value for the TTL field in datagrams.

The default value can be changed to accommodate, for example, a larger internet.

MTU

Another IP function that can be configured is the maximum transfer unit (MTU),

which is used for frames transmitted on the network segment to which a computer is

attached. The MTU is the largest data payload that can be carried in a frame on the

local network. Usually, the MTU can be determined from the network interface and

from the type of hardware used in the local network. In some circumstances—for

example, in the case of a bridged network with dissimilar hardware technologies—

computers might need to use a smaller MTU than is allowed by the local hardware.

This MTU value can be configured in the IP software.

PMTU

The path maximum transfer unit (PMTU) is used to control the size of segments trans-

mitted by TCP. The PMTU value is selected to avoid fragmentation of datagrams

carrying TCP segments. The IP software determines the PMTU value dynamically by

monitoring reports of datagram fragmentation. When the IP software detects that

datagrams are being fragmented, it reduces the PMTU value until fragmentation no

longer occurs. To ensure that the largest possible PMTU value is discovered, the IP

software periodically probes with larger datagrams and watches for fragmentation.

The details of the PMTU mechanism are controlled through configuration parame-

ters in the IP software.

Summary of IP Software Parameters

Table 4.1 lists all the parameters that can be configured in the IP software. This list

was derived from several sources, including the Host Requirements documents (RFC

1122, RFC 1123, and RFC 1127) and the protocol specifications for PMTU discovery

(RFC 1191) and router discovery (RFC 1256).

The Internet Layer 51

006 3273 CH04 10/3/02 5:01 PM Page 51

NOTE

The Host Requirements documents (RFC 1122, RFC 1123, and RFC 1127) summarize and

analyze all the protocols in the TCP/IP suite. They include clarifications to protocol specifica-

tions from other RFCs, “best practices” and hints to implementers, summaries of important

information about protocols, and interactions among protocols. Anyone implementing stacks,

evaluating products, building applications, or otherwise using TCP/IP in nontrivial ways should

read the Host Requirements documents.

TABLE 4.1 Configurable IP Parameters

Parameter Name Values

IP Parameters, Per Host

Be a router on/off

Nonlocal source routing on/off

Policy filters for nonlocal source routing a list of filters

Maximum datagram reassembly size an integer

Default TTL an integer

PMTU aging timeout an integer

MTU plateau table a list of integers

IP Parameters, Per Interface

IP address a 32-bit integer

Subnet mask a 32-bit integer

MTU an integer

All-subnets-MTU on/off

Broadcast address type 0.0.0.0/255.255.255.255

Perform mask discovery on/off

Be a mask supplier on/off

Perform router discovery on/off

Router solicitation address an IP address

Default routers a list of IP addresses

Static routes, list of:

destination an IP address

destination mask an IP address

type-of-service an integer

first-hop router an IP address

Ignore redirects on/off

PMTU an integer

Perform PMTU discovery an integer

CHAPTER 4 Configuring TCP/IP Stacks52

006 3273 CH04 10/3/02 5:01 PM Page 52

The Transport Layer

The two protocols in the transport layer of the TCP/IP protocol suite deliver data

between applications through an internet, using the IP datagram delivery service:

• User Datagram Protocol (UDP) provides best-effort, connectionless delivery of

discrete messages.

• Transmission Control Protocol (TCP) provides reliable, connection-oriented deliv-

ery of arbitrarily long messages or streams of data.

UDP

UDP provides independent delivery of individual messages. The UDP software

accepts outgoing messages from the application layer and adds a header containing

the destination port, the source port, and a checksum. The UDP software then passes

the message to the IP software, along with the IP address of the destination. The IP

software delivers the datagram to the destination computer. At the destination, the

UDP software uses the destination port number to deliver the datagram to the receiv-

ing program.

Because UDP does not guarantee reliable delivery, the destination sends no acknowl-

edgments or other indications of successful receipt of UDP messages. Reliability in

applications that use UDP is the responsibility of application protocols. For example,

the Trivial File Transfer Protocol (TFTP), which uses UDP, employs a send-and-wait

mechanism in which the receiver sends a reply message to the sender to acknowl-

edge receipt of each message from the sender.

DHCP uses UDP to deliver protocol messages between clients and hosts. UDP has

two specific features that are used by DHCP:

• UDP messages can be broadcast and delivered to every computer on a network

segment rather than to just a single destination computer.

• UDP messages can be transmitted with a source IP address of 0.0.0.0 if the

source computer has not yet been assigned an IP address.

These two features, as described in more detail in Chapter 7, “Transmitting DHCP

Messages,” allow a computer to use UDP to locate and communicate with a DHCP

server before the computer has an IP address.

TCP

In contrast to the best-effort message delivery provided by UDP, TCP uses acknowl-

edgments and retransmission to provide reliable delivery of data that is guaranteed

to be correct and in the correct order, without loss or duplication. An application

The Transport Layer 53

006 3273 CH04 10/3/02 5:01 PM Page 53

hands outgoing data to the TCP software, which splits the data into segments. TCP

then uses IP to deliver these segments to the destination. The TCP software at the

destination reconstructs the original message from the individual segments and

passes the data to the receiving application in the correct order.

A TCP receiver sends acknowledgment messages to inform the sender what data has

been successfully delivered to the receiver. If the sender fails to receive an acknowl-

edgment, it retransmits the data until the data is successfully delivered and the

sender receives the acknowledgment.

TCP also includes a form of flow control that enables a receiver to slow down the rate

at which the sender transmits data. The receiver defines a receive window, and the

sender transmits data only within that window. The receiver informs the sender of

the size of its receive window through a field in the TCP header. If the receiver wants

to pause the transmission without shutting down the connection, it sets the receive

window to zero. Later, to resume transmission, the receiver sets the receive window

to a nonzero value.

The Application Layer

The application layer of the TCP/IP protocol suite includes specific protocols that

define interactions between application programs. These application protocols use

the transport layer to exchange messages, which are formatted according to the rules

of the application protocol. Most application protocols are specific to an application;

only a few application protocols are shared among different applications.

Some protocols that are required or crucial to the operation of a TCP/IP intranet are

application protocols. The Domain Name System (DNS), through which human-

friendly names such as

desktop1.genericstartup.com are translated into IP

addresses, uses an application protocol that is carried by either UDP or TCP. As

mentioned earlier in this chapter, in the section “UDP,” TFTP, which is used to trans-

mit software to diskless systems, is an application protocol that uses UDP. Routing

protocols, such as Routing Information Protocol (RIP) and Open Shortest Path First

(OSPF), are carried in UDP and TCP, respectively.

The Client/Server Model

Most protocols in the application layer are based on the client/server model, in which

client applications contact a server to perform application-specific functions. The

server application starts first, and it waits for incoming messages from client applica-

tions. Clients then send messages to the server with requests for some function or

data. DHCP is based on this client/server model.

CHAPTER 4 Configuring TCP/IP Stacks54

006 3273 CH04 10/3/02 5:01 PM Page 54

In a client/server system, the client is configured with the IP address of the server

and contacts the server through that address when required by the application. A

DHCP server can inform a DHCP client of the addresses of servers for many different

application protocols, including DNS, Network Time Protocol (NTP), Simple Mail

Transport Protocol (SMTP), Post Office Protocol (POP), and Network News Transport

Protocol (NNTP).

Summary

The TCP/IP protocol suite is the basis for network communications on the Internet.

TCP/IP includes several layers of protocols, each of which implements specific

services and which, taken together, provide communication between application

programs. Two major styles of communication exist: best-effort datagram delivery

through UDP and reliable, data-stream delivery through TCP. Both TCP and UDP

depend on IP for end-to-end message delivery through an internet.

To accommodate the widest possible variety of network hardware and computer

systems, the protocols in the TCP/IP suite have parameters that can be configured.

Some of these parameters are required for correct operation of the protocols, and

others enhance the performance of applications that use TCP/IP. Configuration of

the most fundamental of these parameters, the IP address, is crucial because incorrect

assignment to one computer could affect the operation of other computers as well.

Summary 55

006 3273 CH04 10/3/02 5:01 PM Page 55