8. The DHCPv4 Server

8.1. Starting and Stopping the DHCPv4 Server

It is recommended that the Kea DHCPv4 server be started and stopped using keactrl (described in Managing Kea with keactrl); however, it is also possible to run the server directly via the kea-dhcp4 command, which accepts the following command-line switches:

  • -c file - specifies the configuration file. This is the only mandatory switch.

  • -d - specifies whether the server logging should be switched to debug/verbose mode. In verbose mode, the logging severity and debuglevel specified in the configuration file are ignored; "debug" severity and the maximum debuglevel (99) are assumed. The flag is convenient for temporarily switching the server into maximum verbosity, e.g. when debugging.

  • -p server-port - specifies the local UDP port on which the server listens. This is only useful during testing, as a DHCPv4 server listening on ports other than the standard ones is not able to handle regular DHCPv4 queries.

  • -P client-port - specifies the remote UDP port to which the server sends all responses. This is only useful during testing, as a DHCPv4 server sending responses to ports other than the standard ones is not able to handle regular DHCPv4 queries.

  • -t file - specifies a configuration file to be tested. kea-dhcp4 loads it, checks it, and exits. During the test, log messages are printed to standard output and error messages to standard error. The result of the test is reported through the exit code (0 = configuration looks OK, 1 = error encountered). The check is not comprehensive; certain checks are possible only when running the server.

  • -T file - specifies a configuration file to be tested. kea-dhcp4 loads it, checks it, and exits. It performs extra checks beyond what -t offers, such as establishing database connections (for the lease backend, host reservations backend, configuration backend, and forensic logging backend), loading hook libraries, parsing hook-library configurations, etc. It does not open UNIX or TCP/UDP sockets, nor does it open or rotate files, as any of these actions could interfere with a running process on the same machine.

  • -v - displays the Kea version and exits.

  • -V - displays the Kea extended version with additional parameters and exits. The listing includes the versions of the libraries dynamically linked to Kea.

  • -W - displays the Kea configuration report and exits. The report is a copy of the config.report file produced by ./configure; it is embedded in the executable binary.

    The contents of the config.report file may also be accessed by examining certain libraries in the installation tree or in the source tree.

    # from installation using libkea-process.so
    $ strings ${prefix}/lib/libkea-process.so | sed -n 's/;;;; //p'
    
    # from sources using libkea-process.so
    $ strings src/lib/process/.libs/libkea-process.so | sed -n 's/;;;; //p'
    
    # from sources using libkea-process.a
    $ strings src/lib/process/.libs/libkea-process.a | sed -n 's/;;;; //p'
    
    # from sources using libcfgrpt.a
    $ strings src/lib/process/cfgrpt/.libs/libcfgrpt.a | sed -n 's/;;;; //p'
    

On startup, the server detects available network interfaces and attempts to open UDP sockets on all interfaces listed in the configuration file. Since the DHCPv4 server opens privileged ports, it requires root access; this daemon must be run as root.

During startup, the server attempts to create a PID file of the form: [runstatedir]/kea/[conf name].kea-dhcp4.pid, where:

  • runstatedir: The value as passed into the build configure script; it defaults to /usr/local/var/run. Note that this value may be overridden at runtime by setting the environment variable KEA_PIDFILE_DIR, although this is intended primarily for testing purposes.
  • conf name: The configuration file name used to start the server, minus all preceding paths and the file extension. For example, given a pathname of /usr/local/etc/kea/myconf.txt, the portion used would be myconf.

If the file already exists and contains the PID of a live process, the server issues a DHCP4_ALREADY_RUNNING log message and exits. It is possible, though unlikely, that the file is a remnant of a system crash and the process to which the PID belongs is unrelated to Kea. In such a case, it would be necessary to manually delete the PID file.

The server can be stopped using the kill command. When running in a console, the server can also be shut down by pressing Ctrl-c. Kea detects the key combination and shuts down gracefully.

The reconfiguration of each Kea server is triggered by the SIGHUP signal. When a server receives the SIGHUP signal it rereads its configuration file and, if the new configuration is valid, uses the new configuration. If the new configuration proves to be invalid, the server retains its current configuration; however, in some cases a fatal error message is logged indicating that the server is no longer providing any service: a working configuration must be loaded as soon as possible.

8.2. DHCPv4 Server Configuration

8.2.1. Introduction

This section explains how to configure the Kea DHCPv4 server using a configuration file.

Before DHCPv4 is started, its configuration file must be created. The basic configuration is as follows:

{
# DHCPv4 configuration starts on the next line
"Dhcp4": {

# First we set up global values
    "valid-lifetime": 4000,
    "renew-timer": 1000,
    "rebind-timer": 2000,

# Next we set up the interfaces to be used by the server.
    "interfaces-config": {
        "interfaces": [ "eth0" ]
    },

# And we specify the type of lease database
    "lease-database": {
        "type": "memfile",
        "persist": true,
        "name": "/var/lib/kea/dhcp4.leases"
    },

# Finally, we list the subnets from which we will be leasing addresses.
    "subnet4": [
        {
            "id": 1,
            "subnet": "192.0.2.0/24",
            "pools": [
                {
                    "pool": "192.0.2.1 - 192.0.2.200"
                }
            ]
        }
    ]
# DHCPv4 configuration ends with the next line
}

}

The following paragraphs provide a brief overview of the parameters in the above example, along with their format. Subsequent sections of this chapter go into much greater detail for these and other parameters.

The lines starting with a hash (#) are comments and are ignored by the server; they do not impact its operation in any way.

The configuration starts in the first line with the initial opening curly bracket (or brace). Each configuration must contain an object specifying the configuration of the Kea module using it. In the example above, this object is called Dhcp4.

The Dhcp4 configuration starts with the "Dhcp4": { line and ends with the corresponding closing brace (in the above example, the brace after the last comment). Everything defined between those lines is considered to be the Dhcp4 configuration.

In general, the order in which those parameters appear does not matter, but there are two caveats. The first one is that the configuration file must be well-formed JSON, meaning that the parameters for any given scope must be separated by a comma, and there must not be a comma after the last parameter. When reordering a configuration file, moving a parameter to or from the last position in a given scope may also require moving the comma. The second caveat is that it is uncommon — although legal JSON — to repeat the same parameter multiple times. If that happens, the last occurrence of a given parameter in a given scope is used, while all previous instances are ignored. This is unlikely to cause any confusion as there are no real-life reasons to keep multiple copies of the same parameter in the configuration file.

The first few DHCPv4 configuration elements define some global parameters. valid-lifetime defines how long the addresses (leases) given out by the server are valid; the default is for a client to be allowed to use a given address for 4000 seconds. (Note that integer numbers are specified as is, without any quotes around them.) renew-timer and rebind-timer are values (also in seconds) that define the T1 and T2 timers that govern when the client begins the renewal and rebind processes.

Note

The lease valid lifetime is expressed as a triplet with minimum, default, and maximum values using configuration entries min-valid-lifetime, valid-lifetime, and max-valid-lifetime. Since Kea 1.9.5, these values may be specified in client classes. The procedure the server uses to select which lifetime value to use is as follows:

If the client query is a BOOTP query, the server always uses the infinite lease time (e.g. 0xffffffff). Otherwise, the server must determine which configured triplet to use by first searching all classes assigned to the query, and then the subnet selected for the query.

Classes are searched in the order they were assigned to the query; the server uses the triplet from the first class that specifies it. If no classes specify the triplet, the server uses the triplet specified by the subnet selected for the client. If the subnet does not explicitly specify it, the server next looks at the subnet's shared-network (if one exists), then for a global specification, and finally the global default.

If the client requested a lifetime value via DHCP option 51, then the lifetime value used is the requested value bounded by the configured triplet. In other words, if the requested lifetime is less than the configured minimum, the configured minimum is used; if it is more than the configured maximum, the configured maximum is used. If the client did not provide a requested value, the lifetime value used is the triplet default value.

Note

Both renew-timer and rebind-timer are optional. The server only sends rebind-timer to the client, via DHCPv4 option code 59, if it is less than valid-lifetime; and it only sends renew-timer, via DHCPv4 option code 58, if it is less than rebind-timer (or valid-lifetime if rebind-timer was not specified). In their absence, the client should select values for T1 and T2 timers according to RFC 2131. See section Sending T1 (Option 58) and T2 (Option 59) for more details on generating T1 and T2.

The interfaces-config map specifies the network interfaces on which the server should listen to DHCP messages. The interfaces parameter specifies a list of network interfaces on which the server should listen. Lists are opened and closed with square brackets, with elements separated by commas. To listen on two interfaces, the interfaces-config element should look like this:

{
"interfaces-config": {
    "interfaces": [ "eth0", "eth1" ]
},
...
}

The next lines define the lease database, the place where the server stores its lease information. This particular example tells the server to use memfile, which is the simplest and fastest database backend. It uses an in-memory database and stores leases on disk in a CSV (comma-separated values) file. This is a very simple configuration example; usually the lease database configuration is more extensive and contains additional parameters. Note that lease-database is an object and opens up a new scope, using an opening brace. Its parameters (just one in this example: type) follow. If there were more than one, they would be separated by commas. This scope is closed with a closing brace. As more parameters for the Dhcp4 definition follow, a trailing comma is present.

Finally, we need to define a list of IPv4 subnets. This is the most important DHCPv4 configuration structure, as the server uses that information to process clients' requests. It defines all subnets from which the server is expected to receive DHCP requests. The subnets are specified with the subnet4 parameter. It is a list, so it starts and ends with square brackets. Each subnet definition in the list has several attributes associated with it, so it is a structure and is opened and closed with braces. At a minimum, a subnet definition must have at least two parameters: subnet, which defines the whole subnet; and pools, which is a list of dynamically allocated pools that are governed by the DHCP server.

The example contains a single subnet. If more than one were defined, additional elements in the subnet4 parameter would be specified and separated by commas. For example, to define three subnets, the following syntax would be used:

{
"subnet4": [
    {
        "id": 1,
        "pools": [ { "pool":  "192.0.2.1 - 192.0.2.200" } ],
        "subnet": "192.0.2.0/24"
    },
    {
        "id": 2,
        "pools": [ { "pool": "192.0.3.100 - 192.0.3.200" } ],
        "subnet": "192.0.3.0/24"
    },
    {
        "id": 3,
        "pools": [ { "pool": "192.0.4.1 - 192.0.4.254" } ],
        "subnet": "192.0.4.0/24"
    }
],
...
}

Note that indentation is optional and is used for aesthetic purposes only. In some cases it may be preferable to use more compact notation.

After all the parameters have been specified, there are two contexts open: global and Dhcp4; thus, two closing curly brackets must be used to close them.

8.2.2. Lease Storage

All leases issued by the server are stored in the lease database. There are three database backends available: memfile (the default), MySQL, PostgreSQL.

8.2.2.1. Memfile - Basic Storage for Leases

The server is able to store lease data in different repositories. Larger deployments may elect to store leases in a database; Lease Database Configuration describes this option. In typical smaller deployments, though, the server stores lease information in a CSV file rather than a database. As well as requiring less administration, an advantage of using a file for storage is that it eliminates a dependency on third-party database software.

The configuration of the memfile backend is controlled through the Dhcp4/lease-database parameters. The type parameter is mandatory and specifies which storage for leases the server should use, through the "memfile" value. The following list gives additional optional parameters that can be used to configure the memfile backend.

  • persist: controls whether the new leases and updates to existing leases are written to the file. It is strongly recommended that the value of this parameter be set to true at all times during the server's normal operation. Not writing leases to disk means that if a server is restarted (e.g. after a power failure), it will not know which addresses have been assigned. As a result, it may assign new clients addresses that are already in use. The value of false is mostly useful for performance-testing purposes. The default value of the persist parameter is true, which enables writing lease updates to the lease file.
  • name: specifies an absolute location of the lease file in which new leases and lease updates are recorded. The default value for this parameter is "[kea-install-dir]/var/lib/kea/kea-leases4.csv".
  • lfc-interval: specifies the interval, in seconds, at which the server will perform a lease file cleanup (LFC). This removes redundant (historical) information from the lease file and effectively reduces the lease file size. The cleanup process is described in more detail later in this section. The default value of the lfc-interval is 3600. A value of 0 disables the LFC.
  • max-row-errors: specifies the number of row errors before the server stops attempting to load a lease file. When the server loads a lease file, it is processed row by row, each row containing a single lease. If a row is flawed and cannot be processed correctly the server logs it, discards the row, and goes on to the next row. This parameter can be used to set a limit on the number of such discards that can occur, after which the server abandons the effort and exits. The default value of 0 disables the limit and allows the server to process the entire file, regardless of how many rows are discarded.

An example configuration of the memfile backend is presented below:

"Dhcp4": {
    "lease-database": {
        "type": "memfile",
        "persist": true,
        "name": "/tmp/kea-leases4.csv",
        "lfc-interval": 1800,
        "max-row-errors": 100
    }
}

This configuration selects /tmp/kea-leases4.csv as the storage for lease information and enables persistence (writing lease updates to this file). It also configures the backend to perform a periodic cleanup of the lease file every 1800 seconds (30 minutes) and sets the maximum number of row errors to 100.

8.2.2.2. Why Is Lease File Cleanup Necessary?

It is important to know how the lease file contents are organized to understand why the periodic lease file cleanup is needed. Every time the server updates a lease or creates a new lease for a client, the new lease information must be recorded in the lease file. For performance reasons, the server does not update the existing client's lease in the file, as this would potentially require rewriting the entire file. Instead, it simply appends the new lease information to the end of the file; the previous lease entries for the client are not removed. When the server loads leases from the lease file, e.g. at server startup, it assumes that the latest lease entry for the client is the valid one. Previous entries are discarded, meaning that the server can reconstruct accurate information about the leases even though there may be many lease entries for each client. However, storing many entries for each client results in a bloated lease file and impairs the performance of the server's startup and reconfiguration, as it needs to process a larger number of lease entries.

Lease file cleanup (LFC) removes all previous entries for each client and leaves only the latest ones. The interval at which the cleanup is performed is configurable, and it should be selected according to the frequency of lease renewals initiated by the clients. The more frequent the renewals, the smaller the value of lfc-interval should be. Note, however, that the LFC takes time and thus it is possible (although unlikely) that, if the lfc-interval is too short, a new cleanup may be started while the previous one is still running. The server would recover from this by skipping the new cleanup when it detected that the previous cleanup was still in progress, but it implies that the actual cleanups will be triggered more rarely than the configured interval. Moreover, triggering a new cleanup adds overhead to the server, which is not able to respond to new requests for a short period of time when the new cleanup process is spawned. Therefore, it is recommended that the lfc-interval value be selected in a way that allows the LFC to complete the cleanup before a new cleanup is triggered.

Lease file cleanup is performed by a separate process (in the background) to avoid a performance impact on the server process. To avoid conflicts between two processes using the same lease files, the LFC process starts with Kea opening a new lease file; the actual LFC process operates on the lease file that is no longer used by the server. There are also other files created as a side effect of the lease file cleanup. The detailed description of the LFC process is located later in this Kea Administrator's Reference Manual: The LFC Process.

8.2.2.3. Lease Database Configuration

Note

Lease database access information must be configured for the DHCPv4 server, even if it has already been configured for the DHCPv6 server. The servers store their information independently, so each server can use a separate database or both servers can use the same database.

Note

Kea requires the database timezone to match the system timezone. For more details, see First-Time Creation of the MySQL Database and First-Time Creation of the PostgreSQL Database.

Lease database configuration is controlled through the Dhcp4/lease-database parameters. The database type must be set to memfile, mysql or postgresql, e.g.:

"Dhcp4": { "lease-database": { "type": "mysql", ... }, ... }

Next, the name of the database to hold the leases must be set; this is the name used when the database was created (see First-Time Creation of the MySQL Database or First-Time Creation of the PostgreSQL Database).

For MySQL or PostgreSQL:

"Dhcp4": { "lease-database": { "name": "database-name" , ... }, ... }

If the database is located on a different system from the DHCPv4 server, the database host name must also be specified:

"Dhcp4": { "lease-database": { "host": "remote-host-name", ... }, ... }

Normally, the database is on the same machine as the DHCPv4 server. In this case, set the value to the empty string:

"Dhcp4": { "lease-database": { "host" : "", ... }, ... }

Should the database use a port other than the default, it may be specified as well:

"Dhcp4": { "lease-database": { "port" : 12345, ... }, ... }

Should the database be located on a different system, the administrator may need to specify a longer interval for the connection timeout:

"Dhcp4": { "lease-database": { "connect-timeout" : timeout-in-seconds, ... }, ... }

The default value of five seconds should be more than adequate for local connections. If a timeout is given, though, it should be an integer greater than zero.

The maximum number of times the server automatically attempts to reconnect to the lease database after connectivity has been lost may be specified:

"Dhcp4": { "lease-database": { "max-reconnect-tries" : number-of-tries, ... }, ... }

If the server is unable to reconnect to the database after making the maximum number of attempts, the server will exit. A value of 0 (the default) disables automatic recovery and the server will exit immediately upon detecting a loss of connectivity (MySQL and PostgreSQL only).

The number of milliseconds the server waits between attempts to reconnect to the lease database after connectivity has been lost may also be specified:

"Dhcp4": { "lease-database": { "reconnect-wait-time" : number-of-milliseconds, ... }, ... }

The default value for MySQL and PostgreSQL is 0, which disables automatic recovery and causes the server to exit immediately upon detecting the loss of connectivity.

"Dhcp4": { "lease-database": { "on-fail" : "stop-retry-exit", ... }, ... }

The possible values are:

  • stop-retry-exit - disables the DHCP service while trying to automatically recover lost connections. Shuts down the server on failure after exhausting max-reconnect-tries. This is the default value for MySQL and PostgreSQL.
  • serve-retry-exit - continues the DHCP service while trying to automatically recover lost connections. Shuts down the server on failure after exhausting max-reconnect-tries.
  • serve-retry-continue - continues the DHCP service and does not shut down the server even if the recovery fails.

Note

Automatic reconnection to database backends is configured individually per backend; this allows users to tailor the recovery parameters to each backend they use. We suggest that users enable it either for all backends or none, so behavior is consistent.

Losing connectivity to a backend for which reconnection is disabled results (if configured) in the server shutting itself down. This includes cases when the lease database backend and the hosts database backend are connected to the same database instance.

It is highly recommended not to change the stop-retry-exit default setting for the lease manager, as it is critical for the connection to be active while processing DHCP traffic. Change this only if the server is used exclusively as a configuration tool.

The host parameter is used by the MySQL and PostgreSQL backends.

Finally, the credentials of the account under which the server will access the database should be set:

"Dhcp4": {
    "lease-database": {
        "user": "user-name",
        "password": "password",
        ...
    },
    ...
}

If there is no password to the account, set the password to the empty string "". (This is the default.)

8.2.2.4. Tuning Database Timeouts

In rare cases, reading or writing to the database may hang. This can be caused by a temporary network issue, or by misconfiguration of the proxy server switching the connection between different database instances. These situations are rare, but users have reported that Kea sometimes hangs while performing database IO operations. Setting appropriate timeout values can mitigate such issues.

MySQL exposes two distinct connection options to configure the read and write timeouts. Kea's corresponding read-timeout and write-timeout configuration parameters specify the timeouts in seconds. For example:

"Dhcp4": { "lease-database": { "read-timeout" : 10, "write-timeout": 20, ... }, ... }

Setting these parameters to 0 is equivalent to not specifying them, and causes the Kea server to establish a connection to the database with the MySQL defaults. In this case, Kea waits indefinitely for the completion of the read and write operations.

MySQL versions earlier than 5.6 do not support setting timeouts for read and write operations. Moreover, the read-timeout and write-timeout parameters can only be specified for the MySQL backend; setting them for any other backend database type causes a configuration error.

To set a timeout in seconds for PostgreSQL, use the tcp-user-timeout parameter. For example:

"Dhcp4": { "lease-database": { "tcp-user-timeout" : 10, ... }, ... }

Specifying this parameter for other backend types causes a configuration error.

Note

The timeouts described here are only effective for TCP connections. Please note that the MySQL client library used by the Kea servers typically connects to the database via a UNIX domain socket when the host parameter is localhost, but establishes a TCP connection for 127.0.0.1.

8.2.3. Hosts Storage

Kea is also able to store information about host reservations in the database. The hosts database configuration uses the same syntax as the lease database. In fact, the Kea server opens independent connections for each purpose, be it lease or hosts information, which gives the most flexibility. Kea can keep leases and host reservations separately, but can also point to the same database. Currently the supported hosts database types are MySQL and PostgreSQL.

The following configuration can be used to configure a connection to MySQL:

"Dhcp4": {
    "hosts-database": {
        "type": "mysql",
        "name": "kea",
        "user": "kea",
        "password": "secret123",
        "host": "localhost",
        "port": 3306
    }
}

Depending on the database configuration, many of the parameters may be optional.

Please note that usage of hosts storage is optional. A user can define all host reservations in the configuration file, and that is the recommended way if the number of reservations is small. However, when the number of reservations grows, it is more convenient to use host storage. Please note that both storage methods (the configuration file and one of the supported databases) can be used together. If hosts are defined in both places, the definitions from the configuration file are checked first and external storage is checked later, if necessary.

Host information can be placed in multiple stores. Operations are performed on the stores in the order they are defined in the configuration file, although this leads to a restriction in ordering in the case of a host reservation addition; read-only stores must be configured after a (required) read-write store, or the addition will fail.

Note

Kea requires the database timezone to match the system timezone. For more details, see First-Time Creation of the MySQL Database and First-Time Creation of the PostgreSQL Database.

8.2.3.1. DHCPv4 Hosts Database Configuration

Hosts database configuration is controlled through the Dhcp4/hosts-database parameters. If enabled, the type of database must be set to mysql or postgresql.

"Dhcp4": { "hosts-database": { "type": "mysql", ... }, ... }

Next, the name of the database to hold the reservations must be set; this is the name used when the lease database was created (see Supported Backends for instructions on how to set up the desired database type):

"Dhcp4": { "hosts-database": { "name": "database-name" , ... }, ... }

If the database is located on a different system than the DHCPv4 server, the database host name must also be specified:

"Dhcp4": { "hosts-database": { "host": remote-host-name, ... }, ... }

Normally, the database is on the same machine as the DHCPv4 server. In this case, set the value to the empty string:

"Dhcp4": { "hosts-database": { "host" : "", ... }, ... }

Should the database use a port different than the default, it may be specified as well:

"Dhcp4": { "hosts-database": { "port" : 12345, ... }, ... }

The maximum number of times the server automatically attempts to reconnect to the host database after connectivity has been lost may be specified:

"Dhcp4": { "hosts-database": { "max-reconnect-tries" : number-of-tries, ... }, ... }

If the server is unable to reconnect to the database after making the maximum number of attempts, the server will exit. A value of 0 (the default) disables automatic recovery and the server will exit immediately upon detecting a loss of connectivity (MySQL and PostgreSQL only).

The number of milliseconds the server waits between attempts to reconnect to the host database after connectivity has been lost may also be specified:

"Dhcp4": { "hosts-database": { "reconnect-wait-time" : number-of-milliseconds, ... }, ... }

The default value for MySQL and PostgreSQL is 0, which disables automatic recovery and causes the server to exit immediately upon detecting the loss of connectivity.

"Dhcp4": { "hosts-database": { "on-fail" : "stop-retry-exit", ... }, ... }

The possible values are:

  • stop-retry-exit - disables the DHCP service while trying to automatically recover lost connections. Shuts down the server on failure after exhausting max-reconnect-tries. This is the default value for MySQL and PostgreSQL.
  • serve-retry-exit - continues the DHCP service while trying to automatically recover lost connections. Shuts down the server on failure after exhausting max-reconnect-tries.
  • serve-retry-continue - continues the DHCP service and does not shut down the server even if the recovery fails.

Note

Automatic reconnection to database backends is configured individually per backend. This allows users to tailor the recovery parameters to each backend they use. We suggest that users enable it either for all backends or none, so behavior is consistent.

Losing connectivity to a backend for which reconnection is disabled results (if configured) in the server shutting itself down. This includes cases when the lease database backend and the hosts database backend are connected to the same database instance.

Finally, the credentials of the account under which the server will access the database should be set:

"Dhcp4": {
    "hosts-database": {
        "user": "user-name",
        "password": "password",
        ...
    },
    ...
}

If there is no password to the account, set the password to the empty string "". (This is the default.)

The multiple-storage extension uses a similar syntax; a configuration is placed into a hosts-databases list instead of into a hosts-database entry, as in:

"Dhcp4": { "hosts-databases": [ { "type": "mysql", ... }, ... ], ... }

If the same host is configured both in-file and in-database, Kea does not issue a warning, as it would if both were specified in the same data source. Instead, the host configured in-file has priority over the one configured in-database.

8.2.3.2. Using Read-Only Databases for Host Reservations With DHCPv4

In some deployments, the user whose name is specified in the database backend configuration may not have write privileges to the database. This is often required by the policy within a given network to secure the data from being unintentionally modified. In many cases administrators have deployed inventory databases, which contain substantially more information about the hosts than just the static reservations assigned to them. The inventory database can be used to create a view of a Kea hosts database and such a view is often read-only.

Kea host-database backends operate with an implicit configuration to both read from and write to the database. If the user does not have write access to the host database, the backend will fail to start and the server will refuse to start (or reconfigure). However, if access to a read-only host database is required for retrieving reservations for clients and/or assigning specific addresses and options, it is possible to explicitly configure Kea to start in "read-only" mode. This is controlled by the readonly boolean parameter as follows:

"Dhcp4": { "hosts-database": { "readonly": true, ... }, ... }

Setting this parameter to false configures the database backend to operate in "read-write" mode, which is also the default configuration if the parameter is not specified.

Note

The readonly parameter is only supported for MySQL and PostgreSQL databases.

8.2.3.3. Tuning Database Timeouts for Hosts Storage

See Tuning Database Timeouts.

8.2.4. Interface Configuration

The DHCPv4 server must be configured to listen on specific network interfaces. The simplest network interface configuration tells the server to listen on all available interfaces:

"Dhcp4": {
    "interfaces-config": {
        "interfaces": [ "*" ]
    },
    ...
}

The asterisk plays the role of a wildcard and means "listen on all interfaces." However, it is usually a good idea to explicitly specify interface names:

"Dhcp4": {
    "interfaces-config": {
        "interfaces": [ "eth1", "eth3" ]
    },
    ...
}

It is possible to use an interface wildcard (*) concurrently with explicit interface names:

"Dhcp4": {
    "interfaces-config": {
        "interfaces": [ "eth1", "eth3", "*" ]
    },
    ...
}

This format should only be used when it is desired to temporarily override a list of interface names and listen on all interfaces.

Some deployments of DHCP servers require that the servers listen on interfaces with multiple IPv4 addresses configured. In these situations, the address to use can be selected by appending an IPv4 address to the interface name in the following manner:

"Dhcp4": {
    "interfaces-config": {
        "interfaces": [ "eth1/10.0.0.1", "eth3/192.0.2.3" ]
    },
    ...
}

Should the server be required to listen on multiple IPv4 addresses assigned to the same interface, multiple addresses can be specified for an interface as in the example below:

"Dhcp4": {
    "interfaces-config": {
        "interfaces": [ "eth1/10.0.0.1", "eth1/10.0.0.2" ]
    },
    ...
}

Alternatively, if the server should listen on all addresses for the particular interface, an interface name without any address should be specified.

Kea supports responding to directly connected clients which do not have an address configured. This requires the server to inject the hardware address of the destination into the data-link layer of the packet being sent to the client. The DHCPv4 server uses raw sockets to achieve this, and builds the entire IP/UDP stack for the outgoing packets. The downside of raw socket use, however, is that incoming and outgoing packets bypass the firewalls (e.g. iptables).

Handling traffic on multiple IPv4 addresses assigned to the same interface can be a challenge, as raw sockets are bound to the interface. When the DHCP server is configured to use the raw socket on an interface to receive DHCP traffic, advanced packet filtering techniques (e.g. the BPF) must be used to receive unicast traffic on the desired addresses assigned to the interface. Whether clients use the raw socket or the UDP socket depends on whether they are directly connected (raw socket) or relayed (either raw or UDP socket).

Therefore, in deployments where the server does not need to provision the directly connected clients and only receives the unicast packets from the relay agents, the Kea server should be configured to use UDP sockets instead of raw sockets. The following configuration demonstrates how this can be achieved:

"Dhcp4": {
    "interfaces-config": {
        "interfaces": [ "eth1", "eth3" ],
        "dhcp-socket-type": "udp"
    },
    ...
}

The dhcp-socket-type parameter specifies that the IP/UDP sockets will be opened on all interfaces on which the server listens, i.e. "eth1" and "eth3" in this example. If dhcp-socket-type is set to raw, it configures the server to use raw sockets instead. If the dhcp-socket-type value is not specified, the default value raw is used.

Using UDP sockets automatically disables the reception of broadcast packets from directly connected clients. This effectively means that UDP sockets can be used for relayed traffic only. When using raw sockets, both the traffic from the directly connected clients and the relayed traffic are handled.

Caution should be taken when configuring the server to open multiple raw sockets on the interface with several IPv4 addresses assigned. If the directly connected client sends the message to the broadcast address, all sockets on this link will receive this message and multiple responses will be sent to the client. Therefore, the configuration with multiple IPv4 addresses assigned to the interface should not be used when the directly connected clients are operating on that link. To use a single address on such an interface, the "interface-name/address" notation should be used.

Note

Specifying the value raw as the socket type does not guarantee that raw sockets will be used! The use of raw sockets to handle traffic from the directly connected clients is currently supported on Linux and BSD systems only. If raw sockets are not supported on the particular OS in use, the server issues a warning and fall back to using IP/UDP sockets.

In a typical environment, the DHCP server is expected to send back a response on the same network interface on which the query was received. This is the default behavior. However, in some deployments it is desired that the outbound (response) packets be sent as regular traffic and the outbound interface be determined by the routing tables. This kind of asymmetric traffic is uncommon, but valid. Kea supports a parameter called outbound-interface that controls this behavior. It supports two values: the first one, same-as-inbound, tells Kea to send back the response on the same interface where the query packet was received. This is the default behavior. The second parameter, use-routing, tells Kea to send regular UDP packets and let the kernel's routing table determine the most appropriate interface. This only works when dhcp-socket-type is set to udp. An example configuration looks as follows:

"Dhcp4": {
    "interfaces-config": {
        "interfaces": [ "eth1", "eth3" ],
        "dhcp-socket-type": "udp",
        "outbound-interface": "use-routing"
    },
    ...
}

Interfaces are re-detected at each reconfiguration. This behavior can be disabled by setting the re-detect value to false, for instance:

"Dhcp4": {
    "interfaces-config": {
        "interfaces": [ "eth1", "eth3" ],
        "re-detect": false
    },
    ...
}

Note that interfaces are not re-detected during config-test.

Usually loopback interfaces (e.g. the lo or lo0 interface) are not configured, but if a loopback interface is explicitly configured and IP/UDP sockets are specified, the loopback interface is accepted.

For example, this setup can be used to run Kea in a FreeBSD jail having only a loopback interface, to service a relayed DHCP request:

"Dhcp4": {
    "interfaces-config": {
        "interfaces": [ "lo0" ],
        "dhcp-socket-type": "udp"
    },
    ...
}

Kea binds the service sockets for each interface on startup. If another process is already using a port, then Kea logs the message and suppresses an error. DHCP service runs, but it is unavailable on some interfaces.

The "service-sockets-require-all" option makes Kea require all sockets to be successfully bound. If any opening fails, Kea interrupts the initialization and exits with a non-zero status. (Default is false).

"Dhcp4": {
    "interfaces-config": {
        "interfaces": [ "eth1", "eth3" ],
        "service-sockets-require-all": true
    },
    ...
}

Sometimes, immediate interruption isn't a good choice. The port can be unavailable only temporary. In this case, retrying the opening may resolve the problem. Kea provides two options to specify the retrying: service-sockets-max-retries and service-sockets-retry-wait-time.

The first defines a maximal number of retries that Kea makes to open a socket. The zero value (default) means that the Kea doesn't retry the process.

The second defines a wait time (in milliseconds) between attempts. The default value is 5000 (5 seconds).

"Dhcp4": {
    "interfaces-config": {
        "interfaces": [ "eth1", "eth3" ],
        "service-sockets-max-retries": 5,
        "service-sockets-retry-wait-time": 5000
    },
    ...
}

If "service-sockets-max-retries" is non-zero and "service-sockets-require-all" is false, then Kea retries the opening (if needed) but does not fail if any socket is still not opened.

8.2.5. Issues With Unicast Responses to DHCPINFORM

The use of UDP sockets has certain benefits in deployments where the server receives only relayed traffic; these benefits are mentioned in Interface Configuration. From the administrator's perspective it is often desirable to configure the system's firewall to filter out unwanted traffic, and the use of UDP sockets facilitates this. However, the administrator must also be aware of the implications related to filtering certain types of traffic, as it may impair the DHCP server's operation.

In this section we focus on the case when the server receives the DHCPINFORM message from the client via a relay. According to RFC 2131, the server should unicast the DHCPACK response to the address carried in the ciaddr field. When the UDP socket is in use, the DHCP server relies on the low-level functions of an operating system to build the data link, IP, and UDP layers of the outgoing message. Typically, the OS first uses ARP to obtain the client's link-layer address to be inserted into the frame's header, if the address is not cached from a previous transaction that the client had with the server. When the ARP exchange is successful, the DHCP message can be unicast to the client, using the obtained address.

Some system administrators block ARP messages in their network, which causes issues for the server when it responds to the DHCPINFORM messages because the server is unable to send the DHCPACK if the preceding ARP communication fails. Since the OS is entirely responsible for the ARP communication and then sending the DHCP packet over the wire, the DHCP server has no means to determine that the ARP exchange failed and the DHCP response message was dropped. Thus, the server does not log any error messages when the outgoing DHCP response is dropped. At the same time, all hooks pertaining to the packet-sending operation will be called, even though the message never reaches its destination.

Note that the issue described in this section is not observed when raw sockets are in use, because, in this case, the DHCP server builds all the layers of the outgoing message on its own and does not use ARP. Instead, it inserts the value carried in the chaddr field of the DHCPINFORM message into the link layer.

Server administrators willing to support DHCPINFORM messages via relays should not block ARP traffic in their networks, or should use raw sockets instead of UDP sockets.

8.2.6. IPv4 Subnet Identifier

The subnet identifier (subnet ID) is a unique number associated with a particular subnet. In principle, it is used to associate clients' leases with their respective subnets. The server configuration should contain unique and stable identifiers for all subnets. When a subnet identifier is not specified for a subnet, it is automatically assigned by the configuration mechanism. The identifiers are assigned starting at 1 and are monotonically increased for each subsequent subnet: 1, 2, 3, ....

If there are multiple subnets configured with auto-generated identifiers and one of them is removed, the subnet identifiers may be renumbered. For example: if there are four subnets and the third is removed, the last subnet will be assigned the identifier that the third subnet had before removal. As a result, the leases stored in the lease database for subnet 3 are now associated with subnet 4, something that may have unexpected consequences. It is one of the reasons why auto-generated subnet identifiers are deprecated starting from Kea version 2.4.0.

Note

The auto-generation of the subnet identifiers will be removed in a future release. Starting from Kea 2.4.0, a subnet without an id entry or with the zero value raises a warning at the configuration time.

Note

Subnet IDs must be greater than zero and less than 4294967295.

The following configuration assigns the specified subnet identifier to a newly configured subnet:

"Dhcp4": {
    "subnet4": [
        {
            "subnet": "192.0.2.0/24",
            "id": 1024,
            ...
        }
    ]
}

This identifier will not change for this subnet unless the id parameter is removed or set to 0. The value of 0 forces auto-generation of the subnet identifier.

8.2.7. IPv4 Subnet Prefix

The subnet prefix is the second way to identify a subnet. Kea can accept non-canonical subnet addresses; for instance, this configuration is accepted:

"Dhcp4": {
    "subnet4": [
        {
           "subnet": "192.0.2.1/24",
            ...
        }
    ]
}

This works even if there is another subnet with the "192.0.2.0/24" prefix; only the textual form of subnets are compared to avoid duplicates.

Note

Abuse of this feature can lead to incorrect subnet selection (see How the DHCPv4 Server Selects a Subnet for the Client).

8.2.8. Configuration of IPv4 Address Pools

The main role of a DHCPv4 server is address assignment. For this, the server must be configured with at least one subnet and one pool of dynamic addresses to be managed. For example, assume that the server is connected to a network segment that uses the 192.0.2.0/24 prefix. The administrator of that network decides that addresses from the range 192.0.2.10 to 192.0.2.20 are going to be managed by the DHCPv4 server. Such a configuration can be achieved in the following way:

"Dhcp4": {
    "subnet4": [
        {
            "subnet": "192.0.2.0/24",
            "pools": [
                { "pool": "192.0.2.10 - 192.0.2.20" }
            ],
            ...
        }
    ]
}

Note that subnet is defined as a simple string, but the pools parameter is actually a list of pools; for this reason, the pool definition is enclosed in square brackets, even though only one range of addresses is specified.

Each pool is a structure that contains the parameters that describe a single pool. Currently there is only one parameter, pool, which gives the range of addresses in the pool.

It is possible to define more than one pool in a subnet; continuing the previous example, further assume that 192.0.2.64/26 should also be managed by the server. It could be written as 192.0.2.64 to 192.0.2.127, or it can be expressed more simply as 192.0.2.64/26. Both formats are supported by Dhcp4 and can be mixed in the pool list. For example, the following pools could be defined:

"Dhcp4": {
    "subnet4": [
        {
            "subnet": "192.0.2.0/24",
            "pools": [
                { "pool": "192.0.2.10-192.0.2.20" },
                { "pool": "192.0.2.64/26" }
            ],
            ...
        }
    ],
    ...
}

White space in pool definitions is ignored, so spaces before and after the hyphen are optional. They can be used to improve readability.

The number of pools is not limited, but for performance reasons it is recommended to use as few as possible.

The server may be configured to serve more than one subnet. To add a second subnet, use a command similar to the following:

"Dhcp4": {
    "subnet4": [
        {
            "subnet": "192.0.2.0/24",
            "pools": [ { "pool": "192.0.2.1 - 192.0.2.200" } ],
            ...
        },
        {
            "subnet": "192.0.3.0/24",
            "pools": [ { "pool": "192.0.3.100 - 192.0.3.200" } ],
            ...
        },
        {
            "subnet": "192.0.4.0/24",
            "pools": [ { "pool": "192.0.4.1 - 192.0.4.254" } ],
            ...
        }
    ]
}

When configuring a DHCPv4 server using prefix/length notation, please pay attention to the boundary values. When specifying that the server can use a given pool, it is also able to allocate the first (typically a network address) and the last (typically a broadcast address) address from that pool. In the aforementioned example of pool 192.0.3.0/24, both the 192.0.3.0 and 192.0.3.255 addresses may be assigned as well. This may be invalid in some network configurations. To avoid this, use the min-max notation.

In a subnet whose prefix length is less than 24, users may wish to exclude all addresses ending in .0 and .255 from being dynamically allocated. For instance, in the subnet 10.0.0.0/8, an administrator may wish to exclude 10.x.y.0 and 10.x.y.255 for all values of x and y, even though only 10.0.0.0 and 10.255.255.255 must be excluded according to RFC standards. The exclude-first-last-24 configuration compatibility flag (Kea DHCPv4 Compatibility Configuration Parameters) does this automatically, rather than requiring explicit configuration of many pools or reservations for fake hosts. When true, it applies only to subnets of 24 prefix length or smaller i.e. larger address space; the default is false.

In this case, "exclude" means to skip these addresses in the free address pickup routine of the allocation engine; if a client explicitly requests or has a host reservation for an address in .0 or .255, it will get it.

Note

Here are some liberties and limits to the values that subnets and pools can take in unusual Kea configurations:

Kea configuration case Allowed Comment
Overlapping subnets Yes Administrator should consider how clients are matched to these subnets.
Overlapping pools in one subnet No Startup error: DHCP4_PARSER_FAIL
Overlapping address pools in different subnets Yes Specifying the same address pool in different subnets can be used as an equivalent of the global address pool. In that case, the server can assign addresses from the same range regardless of the client's subnet. If an address from such a pool is assigned to a client in one subnet, the same address will be renewed for this client if it moves to another subnet. Another client in a different subnet will not be assigned an address already assigned to the client in any of the subnets.
Pools not matching the subnet prefix No Startup error: DHCP4_PARSER_FAIL

8.2.9. Sending T1 (Option 58) and T2 (Option 59)

According to RFC 2131, servers should send values for T1 and T2 that are 50% and 87.5% of the lease lifetime, respectively. By default, kea-dhcp4 does not send either value; it can be configured to send values that are either specified explicitly or that are calculated as percentages of the lease time. The server's behavior is governed by a combination of configuration parameters, two of which have already been mentioned. To send specific, fixed values use the following two parameters:

  • renew-timer - specifies the value of T1 in seconds.
  • rebind-timer - specifies the value of T2 in seconds.

The server only sends T2 if it is less than the valid lease time. T1 is only sent if T2 is being sent and T1 is less than T2; or T2 is not being sent and T1 is less than the valid lease time.

Calculating the values is controlled by the following three parameters.

  • calculate-tee-times - when true, T1 and T2 are calculated as percentages of the valid lease time. It defaults to false.
  • t1-percent - the percentage of the valid lease time to use for T1. It is expressed as a real number between 0.0 and 1.0 and must be less than t2-percent. The default value is 0.50, per RFC 2131.
  • t2-percent - the percentage of the valid lease time to use for T2. It is expressed as a real number between 0.0 and 1.0 and must be greater than t1-percent. The default value is .875, per RFC 2131.

Note

In the event that both explicit values are specified and calculate-tee-times is true, the server will use the explicit values. Administrators with a setup where some subnets or shared-networks use explicit values and some use calculated values must not define the explicit values at any level higher than where they will be used. Inheriting them from too high a scope, such as global, will cause them to have explicit values at every level underneath (shared-networks and subnets), effectively disabling calculated values.

8.2.10. Standard DHCPv4 Options

One of the major features of the DHCPv4 server is the ability to provide configuration options to clients. Most of the options are sent by the server only if the client explicitly requests them using the Parameter Request List option. Those that do not require inclusion in the Parameter Request List option are commonly used options, e.g. "Domain Server", and options which require special behavior, e.g. "Client FQDN", which is returned to the client if the client has included this option in its message to the server.

List of standard DHCPv4 options configurable by an administrator comprises the list of the standard DHCPv4 options whose values can be configured using the configuration structures described in this section. This table excludes the options which require special processing and thus cannot be configured with fixed values. The last column of the table indicates which options can be sent by the server even when they are not requested in the Parameter Request List option, and those which are sent only when explicitly requested.

The following example shows how to configure the addresses of DNS servers, which is one of the most frequently used options. Options specified in this way are considered global and apply to all configured subnets.

"Dhcp4": {
    "option-data": [
        {
           "name": "domain-name-servers",
           "code": 6,
           "space": "dhcp4",
           "csv-format": true,
           "data": "192.0.2.1, 192.0.2.2"
        },
        ...
    ]
}

Note that either name or code is required; there is no need to specify both. space has a default value of dhcp4, so this can be skipped as well if a regular (not encapsulated) DHCPv4 option is defined. Finally, csv-format defaults to true, so it too can be skipped, unless the option value is specified as a hexadecimal string. Therefore, the above example can be simplified to:

"Dhcp4": {
    "option-data": [
        {
           "name": "domain-name-servers",
           "data": "192.0.2.1, 192.0.2.2"
        },
        ...
    ]
}

Defined options are added to the response when the client requests them, with a few exceptions which are always added. To enforce the addition of a particular option, set the always-send flag to true as in:

"Dhcp4": {
    "option-data": [
        {
           "name": "domain-name-servers",
           "data": "192.0.2.1, 192.0.2.2",
           "always-send": true
        },
        ...
    ]
}

The effect is the same as if the client added the option code in the Parameter Request List option (or its equivalent for vendor options):

"Dhcp4": {
    "option-data": [
        {
           "name": "domain-name-servers",
           "data": "192.0.2.1, 192.0.2.2",
           "always-send": true
        },
        ...
    ],
    "subnet4": [
        {
           "subnet": "192.0.3.0/24",
           "option-data": [
               {
                   "name": "domain-name-servers",
                   "data": "192.0.3.1, 192.0.3.2"
               },
               ...
           ],
           ...
        },
        ...
    ],
    ...
}

In the example above, the domain-name-servers option respects the global always-send flag and is always added to responses, but for subnet 192.0.3.0/24, the value is taken from the subnet-level option data specification.

Contrary to always-send, if the never-send flag is set to true for a particular option, the server does not add it to the response. The effect is the same as if the client removed the option code in the Parameter Request List option (or its equivalent for vendor options):

"Dhcp4": {
    "option-data": [
        {
           "name": "domain-name-servers",
           "data": "192.0.2.1, 192.0.2.2"
        },
        ...
    ],
    "subnet4": [
        {
           "subnet": "192.0.3.0/24",
           "option-data": [
               {
                   "name": "domain-name-servers",
                   "never-send": true
               },
               ...
           ],
           ...
        },
        ...
    ],
    ...
}

In the example above, the domain-name-servers option is never added to responses on subnet 192.0.3.0/24. never-send has precedence over always-send, so if both are true the option is not added.

Note

The always-send and never-send flags are sticky, meaning they do not follow the usual configuration inheritance rules. Instead, if they are enabled at least once along the configuration inheritance chain, they are applied - even if they are disabled in other places which would normally receive a higher priority. For instance, if one of the flags is enabled in the global scope, but disabled at the subnet level, it is enabled, disregarding the subnet-level setting.

Note

The never-send flag is less powerful than libdhcp_flex_option.so; for instance, it has no effect on options managed by the server itself. Both always-send and never-send have no effect on options which cannot be requested, for instance from a custom space.

The name parameter specifies the option name. For a list of currently supported names, see List of standard DHCPv4 options configurable by an administrator below. The code parameter specifies the option code, which must match one of the values from that list. The next line specifies the option space, which must always be set to dhcp4 as these are standard DHCPv4 options. For other option spaces, including custom option spaces, see Nested DHCPv4 Options (Custom Option Spaces). The next line specifies the format in which the data will be entered; use of CSV (comma-separated values) is recommended. The sixth line gives the actual value to be sent to clients. The data parameter is specified as normal text, with values separated by commas if more than one value is allowed.

Options can also be configured as hexadecimal values. If csv-format is set to false, option data must be specified as a hexadecimal string. The following commands configure the domain-name-servers option for all subnets with the following addresses: 192.0.3.1 and 192.0.3.2. Note that csv-format is set to false.

"Dhcp4": {
    "option-data": [
        {
            "name": "domain-name-servers",
            "code": 6,
            "space": "dhcp4",
            "csv-format": false,
            "data": "C0 00 03 01 C0 00 03 02"
        },
        ...
    ],
    ...
}

Kea supports the following formats when specifying hexadecimal data:

  • Delimited octets - one or more octets separated by either colons or spaces (":" or " "). While each octet may contain one or two digits, we strongly recommend always using two digits. Valid examples are "ab:cd:ef" and "ab cd ef".
  • String of digits - a continuous string of hexadecimal digits with or without a "0x" prefix. Valid examples are "0xabcdef" and "abcdef".

Care should be taken to use proper encoding when using hexadecimal format; Kea's ability to validate data correctness in hexadecimal is limited.

It is also possible to specify data for binary options as a single-quoted text string within double quotes as shown (note that csv-format must be set to false):

"Dhcp4": {
    "option-data": [
        {
            "name": "user-class",
            "code": 77,
            "space": "dhcp4",
            "csv-format": false,
            "data": "'convert this text to binary'"
        },
        ...
    ],
    ...
}

Most of the parameters in the option-data structure are optional and can be omitted in some circumstances, as discussed in Unspecified Parameters for DHCPv4 Option Configuration.

It is possible to specify or override options on a per-subnet basis. If clients connected to most subnets are expected to get the same values of a given option, administrators should use global options. On the other hand, if different values are used in each subnet, it does not make sense to specify global option values; rather, only subnet-specific ones should be set.

The following commands override the global DNS servers option for a particular subnet, setting a single DNS server with address 192.0.2.3:

"Dhcp4": {
    "subnet4": [
        {
            "option-data": [
                {
                    "name": "domain-name-servers",
                    "code": 6,
                    "space": "dhcp4",
                    "csv-format": true,
                    "data": "192.0.2.3"
                },
                ...
            ],
            ...
        },
        ...
    ],
    ...
}

In some cases it is useful to associate some options with an address pool from which a client is assigned a lease. Pool-specific option values override subnet-specific and global option values; it is not possible to prioritize assignment of pool-specific options via the order of pool declarations in the server configuration.

The following configuration snippet demonstrates how to specify the DNS servers option, which is assigned to a client only if the client obtains an address from the given pool:

"Dhcp4": {
    "subnet4": [
        {
            "pools": [
                {
                    "pool": "192.0.2.1 - 192.0.2.200",
                    "option-data": [
                        {
                            "name": "domain-name-servers",
                            "data": "192.0.2.3"
                         },
                         ...
                    ],
                    ...
                },
                ...
            ],
            ...
        },
        ...
    ],
    ...
}

Options can also be specified in class or host-reservation scope. The current Kea options precedence order is (from most important to least): host reservation, pool, subnet, shared network, class, global.

When a data field is a string and that string contains the comma (,; U+002C) character, the comma must be escaped with two backslashes (\\,; U+005C). This double escape is required because both the routine splitting of CSV data into fields and JSON use the same escape character; a single escape (\,) would make the JSON invalid. For example, the string "foo,bar" must be represented as:

"Dhcp4": {
    "subnet4": [
        {
            "pools": [
                {
                    "option-data": [
                        {
                            "name": "boot-file-name",
                            "data": "foo\\,bar"
                        }
                    ]
                },
                ...
            ],
            ...
        },
        ...
    ],
    ...
}

Some options are designated as arrays, which means that more than one value is allowed. For example, the option time-servers allows the specification of more than one IPv4 address, enabling clients to obtain the addresses of multiple NTP servers.

Custom DHCPv4 Options describes the configuration syntax to create custom option definitions (formats). Creation of custom definitions for standard options is generally not permitted, even if the definition being created matches the actual option format defined in the RFCs. However, there is an exception to this rule for standard options for which Kea currently does not provide a definition. To use such options, a server administrator must create a definition as described in Custom DHCPv4 Options in the dhcp4 option space. This definition should match the option format described in the relevant RFC, but the configuration mechanism allows any option format as there is currently no way to validate it.

The currently supported standard DHCPv4 options are listed in the table below. "Name" and "Code" are the values that should be used as a name/code in the option-data structures. "Type" designates the format of the data; the meanings of the various types are given in List of standard DHCP option types.

List of standard DHCPv4 options configurable by an administrator
Name Code Type Array? Returned if not requested?
time-offset 2 int32 false false
routers 3 ipv4-address true true
time-servers 4 ipv4-address true false
name-servers 5 ipv4-address true false
domain-name-servers 6 ipv4-address true true
log-servers 7 ipv4-address true false
cookie-servers 8 ipv4-address true false
lpr-servers 9 ipv4-address true false
impress-servers 10 ipv4-address true false
resource-location-servers 11 ipv4-address true false
boot-size 13 uint16 false false
merit-dump 14 string false false
domain-name 15 fqdn false true
swap-server 16 ipv4-address false false
root-path 17 string false false
extensions-path 18 string false false
ip-forwarding 19 boolean false false
non-local-source-routing 20 boolean false false
policy-filter 21 ipv4-address true false
max-dgram-reassembly 22 uint16 false false
default-ip-ttl 23 uint8 false false
path-mtu-aging-timeout 24 uint32 false false
path-mtu-plateau-table 25 uint16 true false
interface-mtu 26 uint16 false false
all-subnets-local 27 boolean false false
broadcast-address 28 ipv4-address false false
perform-mask-discovery 29 boolean false false
mask-supplier 30 boolean false false
router-discovery 31 boolean false false
router-solicitation-address 32 ipv4-address false false
static-routes 33 ipv4-address true false
trailer-encapsulation 34 boolean false false
arp-cache-timeout 35 uint32 false false
ieee802-3-encapsulation 36 boolean false false
default-tcp-ttl 37 uint8 false false
tcp-keepalive-interval 38 uint32 false false
tcp-keepalive-garbage 39 boolean false false
nis-domain 40 string false false
nis-servers 41 ipv4-address true false
ntp-servers 42 ipv4-address true false
vendor-encapsulated-options 43 empty false false
netbios-name-servers 44 ipv4-address true false
netbios-dd-server 45 ipv4-address true false
netbios-node-type 46 uint8 false false
netbios-scope 47 string false false
font-servers 48 ipv4-address true false
x-display-manager 49 ipv4-address true false
dhcp-option-overload 52 uint8 false false
dhcp-server-identifier 54 ipv4-address false true
dhcp-message 56 string false false
dhcp-max-message-size 57 uint16 false false
vendor-class-identifier 60 string false false
nwip-domain-name 62 string false false
nwip-suboptions 63 binary false false
nisplus-domain-name 64 string false false
nisplus-servers 65 ipv4-address true false
tftp-server-name 66 string false false
boot-file-name 67 string false false
mobile-ip-home-agent 68 ipv4-address true false
smtp-server 69 ipv4-address true false
pop-server 70 ipv4-address true false
nntp-server 71 ipv4-address true false
www-server 72 ipv4-address true false
finger-server 73 ipv4-address true false
irc-server 74 ipv4-address true false
streettalk-server 75 ipv4-address true false
streettalk-directory-assistance-server 76 ipv4-address true false
user-class 77 binary false false
slp-directory-agent 78 record (boolean, ipv4-address) true false
slp-service-scope 79 record (boolean, string) false false
nds-server 85 ipv4-address true false
nds-tree-name 86 string false false
nds-context 87 string false false
bcms-controller-names 88 fqdn true false
bcms-controller-address 89 ipv4-address true false
client-system 93 uint16 true false
client-ndi 94 record (uint8, uint8, uint8) false false
uuid-guid 97 record (uint8, binary) false false
uap-servers 98 string false false
geoconf-civic 99 binary false false
pcode 100 string false false
tcode 101 string false false
v6-only-preferred 108 uint32 false false
netinfo-server-address 112 ipv4-address true false
netinfo-server-tag 113 string false false
v4-captive-portal 114 string false false
auto-config 116 uint8 false false
name-service-search 117 uint16 true false
domain-search 119 fqdn true false
vivco-suboptions 124 record (uint32, binary) false false
vivso-suboptions 125 uint32 false false
pana-agent 136 ipv4-address true false
v4-lost 137 fqdn false false
capwap-ac-v4 138 ipv4-address true false
sip-ua-cs-domains 141 fqdn true false
v4-sztp-redirect 143 tuple true false
rdnss-selection 146 record (uint8, ipv4-address, ipv4-address, fqdn) true false
v4-portparams 159 record (uint8, psid) false false
v4-dnr 162 record (uint16, uint16, uint8, fqdn, binary) false false
option-6rd 212 record (uint8, uint8, ipv6-address, ipv4-address) true false
v4-access-domain 213 fqdn false false

Note

The default-url option was replaced with v4-captive-portal in Kea 2.1.2, as introduced by RFC 8910. The new option has exactly the same format as the old one. The general perception is that default-url was seldom used. If you used it and migrating, please replace default-url with v4-captive-portal and your configuration will continue to work as before.

Kea also supports other options than those listed above; the following options are returned by the Kea engine itself and in general should not be configured manually.

List of standard DHCPv4 options managed by Kea on its own and not directly configurable by an administrator
Name Code Type Description
subnet-mask 1 ipv4-address calculated automatically, based on subnet definition.
host-name 12 string sent by client, generally governed by the DNS configuration.
dhcp-requested-address 50 ipv4-address may be sent by the client and the server should not set it.
dhcp-lease-time 51 uint32 set automatically based on the valid-lifetime parameter.
dhcp-message-type 53 string sent by clients and servers. Set by the Kea engine depending on the situation and should never be configured explicitly.
dhcp-parameter-request-list 55 uint8 array sent by clients and should never be sent by the server.
dhcp-renewal-time 58 uint32 governed by renew-timer parameter.
dhcp-rebinding-time 59 uint32 governed by rebind-timer parameter.
dhcp-client-identifier 61 binary sent by client, echoed back with the value sent by the client.
fqdn 81 record (uint8, uint8, uint8, fqdn) part of the DDNS and D2 configuration.
dhcp-agent-options 82 empty sent by the relay agent. This is an empty container option; see RAI option detail later in this section.
authenticate 90 binary sent by client, Kea does not yet validate it.
client-last-transaction-time 91 uint32 sent by client, server does not set it.
associated-ip 92 ipv4-address array sent by client, server responds with list of addresses.
subnet-selection 118 ipv4-address if present in client's messages, will be used in the subnet selection process.

The following table lists all option types used in the previous two tables with a description of what values are accepted for them.

List of standard DHCP option types
Name Meaning
binary An arbitrary string of bytes, specified as a set of hexadecimal digits.
boolean A boolean value with allowed values true or false.
empty No value; data is carried in sub-options.
fqdn Fully qualified domain name (e.g. www.example.com).
ipv4-address IPv4 address in the usual dotted-decimal notation (e.g. 192.0.2.1).
ipv6-address IPv6 address in the usual colon notation (e.g. 2001:db8::1).
ipv6-prefix IPv6 prefix and prefix length specified using CIDR notation, e.g. 2001:db8:1::/64. This data type is used to represent an 8-bit field conveying a prefix length and the variable length prefix value.
psid PSID and PSID length separated by a slash, e.g. 3/4 specifies PSID=3 and PSID length=4. In the wire format it is represented by an 8-bit field carrying PSID length (in this case equal to 4) and the 16-bits-long PSID value field (in this case equal to "0011000000000000b" using binary notation). Allowed values for a PSID length are 0 to 16. See RFC 7597 for details about the PSID wire representation.
record Structured data that may be comprised of any types (except "record" and "empty"). The array flag applies to the last field only.
string Any text. Please note that Kea silently discards any terminating/trailing nulls from the end of "string" options when unpacking received packets. This is in keeping with RFC 2132, Section 2.
tuple A length encoded as an 8-bit (16-bit for DHCPv6) unsigned integer followed by a string of this length.
uint8 An 8-bit unsigned integer with allowed values 0 to 255.
uint16 A 16-bit unsigned integer with allowed values 0 to 65535.
uint32 A 32-bit unsigned integer with allowed values 0 to 4294967295.
int8 An 8-bit signed integer with allowed values -128 to 127.
int16 A 16-bit signed integer with allowed values -32768 to 32767.
int32 A 32-bit signed integer with allowed values -2147483648 to 2147483647.

Kea also supports the Relay Agent Information (RAI, defined in RFC 3046) option, sometimes referred to as the relay option, agent option, or simply option 82. The option itself is just a container and does not convey any information on its own. The following table contains a list of RAI sub-options that Kea can understand. The RAI and its sub-options are inserted by the relay agent and received by Kea; there is no need for Kea to be configured with those options. Kea's classification and flex-id in host reservations can be used to process those and other options no listed in the table below.

List of RAI sub-options that Kea can understand
Name Code Comment
circuit-id 1 Used when host-reservation-identifiers is set to circuit-id.
remote-id 2 Can be used with flex-id to identify hosts.
link-selection 5 If present, used to select the appropriate subnet.
subscriber-id 6 Can be used with flex-id to identify hosts.
server-id-override 11 If sent by the relay, Kea accepts it as the server-id.
relay-id 12 Identifies the relay
relay-port 19 If sent by the relay, Kea sends back its responses to this port.

All other RAI sub-options (including those not listed here) can be used in client classification to classify incoming packets to specific classes and/or by libdhcp_flex_id.so to construct a unique device identifier. For more information about expressions used in client classification, and flex-id, see Client Classification. The RAI sub-options can be referenced using relay4[option-code].hex. For example, to classify packets based on the remote-id (sub-option code 2), one would use relay4[2].hex. An example client class that would include all packets with a specific remote-id value would looks as follows:

"Dhcp4": {
    "client-classes": [
        {
            "name": "remote-id-1020304",
            "test": "relay4[2].hex == 0x01020304",
            ...
        }
    ],
    ...
}

Classes may be used to segregate traffic into a relatively small number of groups, which then can be used to select specific subnets, pools and extra options and more. If per host behavior is necessary, using host reservations with flex-id is strongly recommended.

8.2.11. Custom DHCPv4 Options

Kea supports custom (non-standard) DHCPv4 options. Let's say that we want to define a new DHCPv4 option called foo, which will have code 222 and will convey a single, unsigned, 32-bit integer value. Such an option can be defined by putting the following entry in the configuration file:

"Dhcp4": {
    "option-def": [
        {
            "name": "foo",
            "code": 222,
            "type": "uint32",
            "array": false,
            "record-types": "",
            "space": "dhcp4",
            "encapsulate": ""
        },
        ...
    ],
    ...
}

The false value of the array parameter determines that the option does NOT comprise an array of uint32 values but is, instead, a single value. Two other parameters have been left blank: record-types and encapsulate. The former specifies the comma-separated list of option data fields, if the option comprises a record of data fields. The record-types value should be non-empty if type is set to "record"; otherwise it must be left blank. The latter parameter specifies the name of the option space being encapsulated by the particular option. If the particular option does not encapsulate any option space, the parameter should be left blank. Note that the option-def configuration statement only defines the format of an option and does not set its value(s).

The name, code, and type parameters are required; all others are optional. The array parameter default value is false. The record-types and encapsulate parameters default values are blank (""). The default space is dhcp4.

Once the new option format is defined, its value is set in the same way as for a standard option. For example, the following commands set a global value that applies to all subnets.

"Dhcp4": {
    "option-data": [
        {
            "name": "foo",
            "code": 222,
            "space": "dhcp4",
            "csv-format": true,
            "data": "12345"
        },
        ...
    ],
    ...
}

New options can take more complex forms than the simple use of primitives (uint8, string, ipv4-address, etc.); it is possible to define an option comprising a number of existing primitives.

For example, say we want to define a new option that will consist of an IPv4 address, followed by an unsigned 16-bit integer, followed by a boolean value, followed by a text string. Such an option could be defined in the following way:

"Dhcp4": {
    "option-def": [
        {
            "name": "bar",
            "code": 223,
            "space": "dhcp4",
            "type": "record",
            "array": false,
            "record-types": "ipv4-address, uint16, boolean, string",
            "encapsulate": ""
        },
        ...
    ],
    ...
}

The type parameter is set to "record" to indicate that the option contains multiple values of different types. These types are given as a comma-separated list in the record-types field and should be ones from those listed in List of standard DHCP option types.

The option's values are set in an option-data statement as follows:

"Dhcp4": {
    "option-data": [
        {
            "name": "bar",
            "space": "dhcp4",
            "code": 223,
            "csv-format": true,
            "data": "192.0.2.100, 123, true, Hello World"
        }
    ],
    ...
}

The csv-format parameter is set to true to indicate that the data field comprises a comma-separated list of values. The values in data must correspond to the types set in the record-types field of the option definition.

When array is set to true and type is set to "record", the last field is an array, i.e. it can contain more than one value, as in:

"Dhcp4": {
    "option-def": [
        {
            "name": "bar",
            "code": 223,
            "space": "dhcp4",
            "type": "record",
            "array": true,
            "record-types": "ipv4-address, uint16",
            "encapsulate": ""
        },
        ...
    ],
    ...
}

The new option content is one IPv4 address followed by one or more 16-bit unsigned integers.

Note

In general, boolean values are specified as true or false, without quotes. Some specific boolean parameters may also accept "true", "false", 0, 1, "0", and "1".

Note

Numbers can be specified in decimal or hexadecimal format. The hexadecimal format can be either plain (e.g. abcd) or prefixed with 0x (e.g. 0xabcd).

8.2.12. DHCPv4 Private Options

Options with a code between 224 and 254 are reserved for private use. They can be defined at the global scope or at the client-class local scope; this allows option definitions to be used depending on context, and option data to be set accordingly. For instance, to configure an old PXEClient vendor:

"Dhcp4": {
    "client-classes": [
        {
            "name": "pxeclient",
            "test": "option[vendor-class-identifier].text == 'PXEClient'",
            "option-def": [
                {
                    "name": "configfile",
                    "code": 209,
                    "type": "string"
                }
            ],
            ...
        },
        ...
    ],
    ...
}

As the Vendor-Specific Information (VSI) option (code 43) has a vendor-specific format, i.e. can carry either raw binary value or sub-options, this mechanism is also available for this option.

In the following example taken from a real configuration, two vendor classes use option 43 for different and incompatible purposes:

"Dhcp4": {
    "option-def": [
        {
            "name": "cookie",
            "code": 1,
            "type": "string",
            "space": "APC"
        },
        {
            "name": "mtftp-ip",
            "code": 1,
            "type": "ipv4-address",
            "space": "PXE"
        },
        ...
    ],
    "client-classes": [
        {
            "name": "APC",
            "test": "option[vendor-class-identifier].text == 'APC'",
            "option-def": [
                {
                    "name": "vendor-encapsulated-options",
                    "type": "empty",
                    "encapsulate": "APC"
                }
            ],
            "option-data": [
                {
                    "name": "cookie",
                    "space": "APC",
                    "data": "1APC"
                },
                {
                    "name": "vendor-encapsulated-options"
                },
                ...
            ],
            ...
        },
        {
            "name": "PXE",
            "test": "option[vendor-class-identifier].text == 'PXE'",
            "option-def": [
                {
                    "name": "vendor-encapsulated-options",
                    "type": "empty",
                    "encapsulate": "PXE"
                }
            ],
            "option-data": [
                {
                    "name": "mtftp-ip",
                    "space": "PXE",
                    "data": "0.0.0.0"
                },
                {
                    "name": "vendor-encapsulated-options"
                },
                ...
            ],
            ...
        },
        ...
    ],
    ...
}

The definition used to decode a VSI option is:

  1. The local definition of a client class the incoming packet belongs to;
  2. If none, the global definition;
  3. If none, the last-resort definition described in the next section, DHCPv4 Vendor-Specific Options (backward-compatible with previous Kea versions).

Note

This last-resort definition for the Vendor-Specific Information option (code 43) is not compatible with a raw binary value. When there are known cases where a raw binary value will be used, a client class must be defined with both a classification expression matching these cases and an option definition for the VSI option with a binary type and no encapsulation.

Note

By default, in the Vendor-Specific Information option (code 43), sub-option code 0 and 255 mean PAD and END respectively, according to RFC 2132. In other words, the sub-option code values of 0 and 255 are reserved. Kea does, however, allow users to define sub-option codes from 0 to 255. If sub-options with codes 0 and/or 255 are defined, bytes with that value are no longer treated as a PAD or an END, but as the sub-option code when parsing a VSI option in an incoming query.

Option 43 input processing (also called unpacking) is deferred so that it happens after classification. This means clients cannot be classified using option 43 sub-options. The definition used to unpack option 43 is determined as follows:

  • If defined at the global scope, this definition is used.
  • If defined at client class scope and the packet belongs to this class, the client class definition is used.
  • If not defined at global scope nor in a client class to which the packet belongs, the built-in last resort definition is used. This definition only says the sub-option space is "vendor-encapsulated-options-space".

The output definition selection is a bit simpler:

  • If the packet belongs to a client class which defines the option 43, use this definition.
  • If defined at the global scope, use this definition.
  • Otherwise, use the built-in last-resort definition.

Since they use a specific/per vendor option space, sub-options are defined at the global scope.

Note

Option definitions in client classes are allowed only for this limited option set (codes 43 and from 224 to 254), and only for DHCPv4.

8.2.13. DHCPv4 Vendor-Specific Options

Currently there are two option spaces defined for kea-dhcp4: dhcp4 (for the top-level DHCPv4 options) and "vendor-encapsulated-options-space", which is empty by default but in which options can be defined. Those options are carried in the Vendor-Specific Information option (code 43). The following examples show how to define an option foo with code 1 that comprises an IPv4 address, an unsigned 16-bit integer, and a string. The foo option is conveyed in a Vendor-Specific Information option.

The first step is to define the format of the option:

"Dhcp4": {
    "option-def": [
        {
            "name": "foo",
            "code": 1,
            "space": "vendor-encapsulated-options-space",
            "type": "record",
            "array": false,
            "record-types": "ipv4-address, uint16, string",
            "encapsulate": ""
        }
    ],
    ...
}

Note that the option space is set to "vendor-encapsulated-options-space". Once the option format is defined, the next step is to define actual values for that option:

"Dhcp4": {
    "option-data": [
        {
            "name": "foo",
            "space": "vendor-encapsulated-options-space",
            "code": 1,
            "csv-format": true,
            "data": "192.0.2.3, 123, Hello World"
        }
    ],
    ...
}

In this example, we also include the Vendor-Specific Information option, which conveys our sub-option foo. This is required; otherwise, the option will not be included in messages sent to the client.

"Dhcp4": {
    "option-data": [
        {
            "name": "vendor-encapsulated-options"
        }
    ],
    ...
}

Alternatively, the option can be specified using its code.

"Dhcp4": {
    "option-data": [
        {
            "code": 43
        }
    ],
    ...
}

Another popular option that is often somewhat imprecisely called the "vendor option" is option 125. Its proper name is the "vendor-independent vendor-specific information option" or "vivso". The idea behind vivso options is that each vendor has its own unique set of options with their own custom formats. The vendor is identified by a 32-bit unsigned integer called enterprise-number or vendor-id.

The standard spaces defined in Kea and their options are:

  • vendor-4491: Cable Television Laboratories, Inc. for DOCSIS3 options:
option code option name option description
1 oro ORO (or Option Request Option), used by clients to request a list of options they are interested in.
2 tftp-servers a list of IPv4 addresses of TFTP servers to be used by the cable modem

In Kea, each vendor is represented by its own vendor space. Since there are hundreds of vendors and they sometimes use different option definitions for different hardware, it is impossible for Kea to support them all natively. Fortunately, it is easy to define support for new vendor options. As an example, the Genexis home gateway device requires the vivso 125 option to be sent with a sub-option 2 that contains a string with the TFTP server URL. To support such a device, three steps are needed: first, establish option definitions that explain how the option is supposed to be formed; second, define option values; and third, tell Kea when to send those specific options, via client classification.

An example snippet of a configuration could look similar to the following:

"Dhcp4": {
    // First, we need to define that the sub-option 2 in vivso option for
    // vendor-id 25167 has a specific format (it's a plain string in this example).
    // After this definition, we can specify values for option tftp.
    "option-def": [
        {
            // We define a short name, so the option can be referenced by name.
            // The option has code 2 and resides within vendor space 25167.
            // Its data is a plain string.
            "name": "tftp",
            "code": 2,
            "space": "vendor-25167",
            "type": "string"
        }
    ],

    "client-classes": [
        {
            // We now need to tell Kea how to recognize when to use vendor space 25167.
            // Usually we can use a simple expression, such as checking if the device
            // sent a vivso option with specific vendor-id, e.g. "vendor[4491].exists".
            // Unfortunately, Genexis is a bit unusual in this aspect, because it
            // doesn't send vivso. In this case we need to look into the vendor class
            // (option code 60) and see if there's a specific string that identifies
            // the device. Alternatively, one can make use of the automated `VENDOR_CLASS_`
            // client class and replace "name" and "test" with `"name": "VENDOR_CLASS_HMC1000"`
            // and no test expression.
            "name": "cpe_genexis",
            "test": "substring(option[60].hex,0,7) == 'HMC1000'",

            // Once the device is recognized, we want to send two options:
            // the vivso option with vendor-id set to 25167, and a sub-option 2.
            "option-data": [
                {
                    "name": "vivso-suboptions",
                    "data": "25167"
                },

                // The sub-option 2 value is defined as any other option. However,
                // we want to send this sub-option 2, even when the client didn't
                // explicitly request it (often there is no way to do that for
                // vendor options). Therefore we use always-send to force Kea
                // to always send this option when 25167 vendor space is involved.
                {
                    "name": "tftp",
                    "space": "vendor-25167",
                    "data": "tftp://192.0.2.1/genexis/HMC1000.v1.3.0-R.img",
                    "always-send": true
                }
            ]
        }
    ]
}

By default, Kea sends back only those options that are requested by a client, unless there are protocol rules that tell the DHCP server to always send an option. This approach works nicely in most cases and avoids problems with clients refusing responses with options they do not understand. However, the situation with vendor options is more complex, as they are not requested the same way as other options, are not well-documented in official RFCs, or vary by vendor.

Some vendors (such as DOCSIS, identified by vendor option 4491) have a mechanism to request specific vendor options and Kea is able to honor those (sub-option 1). Unfortunately, for many other vendors, such as Genexis (25167, discussed above), Kea does not have such a mechanism, so it cannot send any sub-options on its own. To solve this issue, we devised the concept of persistent options. Kea can be told to always send options, even if the client did not request them. This can be achieved by adding "always-send": true to the option data entry. Note that in this particular case an option is defined in vendor space 25167. With always-send enabled, the option is sent every time there is a need to deal with vendor space 25167.

This is also how kea-dhcp4 can be configured to send multiple vendor options from different vendors, along with each of their specific vendor ID. If these options need to be sent by the server regardless of whether the client specified any enterprise number, "always-send": true must be configured for the suboptions that will be included in the vivso-suboptions option (code 125).

"Dhcp4": {
    "option-data": [
        # Typically DHCPv4 clients will send a Parameter Request List option (code 55) for
        # vivso-suboptions (code 125), and that is enough for Kea to understand that it needs to
        # send the option. These options still need to be defined in the configuration, one per
        # each vendor, but they don't need "always-send" enabled in that case. For misbehaving
        # clients that to do not explicitly request it, one may alternatively set "always-send"
        # to true for them as well. This is referring to the following two entries in option-data.
        {
            "name": "vivso-suboptions",
            "space": "dhcp4",
            "data": "2234"
        },
        {
            "name": "vivso-suboptions",
            "space": "dhcp4",
            "data": "3561"
        },
        {
            "always-send": true,
            "data": "tagged",
            "name": "tag",
            "space": "vendor-2234"
        },
        {
            "always-send": true,
            "data": "https://example.com:1234/path",
            "name": "url",
            "space": "vendor-3561"
        }
    ],
    "option-def": [
        {
            "code": 22,
            "name": "tag",
            "space": "vendor-2234",
            "type": "string"
        },
        {
            "code": 11,
            "name": "url",
            "space": "vendor-3561",
            "type": "string"
        }
    ]
}

Another possibility is to redefine the option; see DHCPv4 Private Options.

Kea comes with several example configuration files. Some of them showcase how to configure options 60 and 43. See doc/examples/kea4/vendor-specific.json and doc/examples/kea4/vivso.json in the Kea sources.

Note

kea-dhcp4 is able to recognize multiple Vendor Class Identifier options (code 60) with different vendor IDs in the client requests and to send multiple vivso options (code 125) in the responses, one for each vendor.

kea-dhcp4 honors DOCSIS sub-option 1 (ORO) and adds only requested options if this sub-option is present in the client request.

Currently only one vendor is supported for the vivco-suboptions (code 124) option. Specifying multiple enterprise numbers within a single option instance or multiple options with different enterprise numbers is not supported.

8.2.14. Nested DHCPv4 Options (Custom Option Spaces)

It is sometimes useful to define a completely new option space, such as when a user creates a new option in the standard option space (dhcp4) and wants this option to convey sub-options. Since they are in a separate space, sub-option codes have a separate numbering scheme and may overlap with the codes of standard options.

Note that the creation of a new option space is not required when defining sub-options for a standard option, because one is created by default if the standard option is meant to convey any sub-options (see DHCPv4 Vendor-Specific Options).

If we want a DHCPv4 option called container with code 222, that conveys two sub-options with codes 1 and 2, we first need to define the new sub-options:

"Dhcp4": {
    "option-def": [
        {
            "name": "subopt1",
            "code": 1,
            "space": "isc",
            "type": "ipv4-address",
            "record-types": "",
            "array": false,
            "encapsulate": ""
        },
        {
            "name": "subopt2",
            "code": 2,
            "space": "isc",
            "type": "string",
            "record-types": "",
            "array": false,
            "encapsulate": ""
        }
    ],
    ...
}

Note that we have defined the options to belong to a new option space (in this case, "isc").

The next step is to define a regular DHCPv4 option with the desired code and specify that it should include options from the new option space:

"Dhcp4": {
    "option-def": [
        {
            "name": "container",
            "code": 222,
            "space": "dhcp4",
            "type": "empty",
            "array": false,
            "record-types": "",
            "encapsulate": "isc"
        },
        ...
    ],
    ...
}

The name of the option space in which the sub-options are defined is set in the encapsulate field. The type field is set to "empty", to indicate that this option does not carry any data other than sub-options.

Finally, we can set values for the new options:

{
  "Dhcp4": {
    "option-data": [
        {
            "name": "subopt1",
            "code": 1,
            "space": "isc",
            "data": "192.0.2.3"
        },
        {
            "name": "subopt2",
            "code": 2,
            "space": "isc",
            "data": "Hello world"
        },
        {
            "name": "container",
            "code": 222,
            "space": "dhcp4"
        }
    ]
  }
}

It is possible to create an option which carries some data in addition to the sub-options defined in the encapsulated option space. For example, if the container option from the previous example were required to carry a uint16 value as well as the sub-options, the type value would have to be set to "uint16" in the option definition. (Such an option would then have the following data structure: DHCP header, uint16 value, sub-options.) The value specified with the data parameter — which should be a valid integer enclosed in quotes, e.g. "123" — would then be assigned to the uint16 field in the container option.

8.2.15. Unspecified Parameters for DHCPv4 Option Configuration

In many cases it is not required to specify all parameters for an option configuration, and the default values can be used. However, it is important to understand the implications of not specifying some of them, as it may result in configuration errors. The list below explains the behavior of the server when a particular parameter is not explicitly specified:

  • name - the server requires either an option name or an option code to identify an option. If this parameter is unspecified, the option code must be specified.
  • code - the server requires either an option name or an option code to identify an option; this parameter may be left unspecified if the name parameter is specified. However, this also requires that the particular option have a definition (either as a standard option or an administrator-created definition for the option using an option-def structure), as the option definition associates an option with a particular name. It is possible to configure an option for which there is no definition (unspecified option format). Configuration of such options requires the use of the option code.
  • space - if the option space is unspecified it defaults to dhcp4, which is an option space holding standard DHCPv4 options.
  • data - if the option data is unspecified it defaults to an empty value. The empty value is mostly used for the options which have no payload (boolean options), but it is legal to specify empty values for some options which carry variable-length data and for which the specification allows a length of 0. For such options, the data parameter may be omitted in the configuration.
  • csv-format - if this value is not specified, the server assumes that the option data is specified as a list of comma-separated values to be assigned to individual fields of the DHCP option.

8.2.16. Support for Long Options

The kea-dhcp4 server partially supports long options (RFC3396). Since Kea 2.1.6, the server accepts configuring long options and sub-options (longer than 255 bytes). The options and sub-options are stored internally in their unwrapped form and they can be processed as usual using the parser language. On send, the server splits long options and sub-options into multiple options and sub-options, using the respective option code.

{
"option-def": [
    {
        "array": false,
        "code": 240,
        "encapsulate": "",
        "name": "my-option",
        "space": "dhcp4",
        "type": "string"
    }
],
"subnet4": [
    {
        "id": 1,
        "subnet": "192.0.2.0/24",
        "reservations": [
            {
                "hw-address": "aa:bb:cc:dd:ee:ff",
                "option-data": [
                    {
                        "always-send": false,
                        "code": 240,
                        "name": "my-option",
                        "csv-format": true,
                        "data": "data \
                                 -00010203040506070809-00010203040506070809-00010203040506070809-00010203040506070809 \
                                 -00010203040506070809-00010203040506070809-00010203040506070809-00010203040506070809 \
                                 -00010203040506070809-00010203040506070809-00010203040506070809-00010203040506070809 \
                                 -data",
                        "space": "dhcp4"
                    }
                ]
            }
        ]
    }
],
...
}

Note

In the example above, the data has been wrapped into several lines for clarity, but Kea does not support wrapping in the configuration file.

This example illustrates configuring a custom long option (exceeding 255 octets) in a reservation. When sending a response, the server splits this option into two options, each with the code 240.

Note

Currently the server does not support storing long options in databases, either host reservations or the configuration backend.

The server is also able to receive packets with split options (options using the same option code) and to fuse the data chunks into one option. This is also supported for sub-options if each sub-option data chunk also contains the sub-option code and sub-option length.

8.2.17. Stateless Configuration of DHCPv4 Clients

The DHCPv4 server supports stateless client configuration, whereby the client has an IP address configured (e.g. using manual configuration) and only contacts the server to obtain other configuration parameters, such as addresses of DNS servers. To obtain the stateless configuration parameters, the client sends the DHCPINFORM message to the server with the ciaddr set to the address that the client is currently using. The server unicasts the DHCPACK message to the client that includes the stateless configuration ("yiaddr" not set).

The server responds to the DHCPINFORM when the client is associated with a subnet defined in the server's configuration. An example subnet configuration looks like this:

"Dhcp4": {
    "subnet4": [
        {
            "id": 1,
            "subnet": "192.0.2.0/24",
            "option-data": [
            {
                "name": "domain-name-servers",
                "code": 6,
                "data": "192.0.2.200,192.0.2.201",
                "csv-format": true,
                "space": "dhcp4"
            }
            ]
        }
    ]
}

This subnet specifies the single option which will be included in the DHCPACK message to the client in response to DHCPINFORM. The subnet definition does not require the address pool configuration if it will be used solely for stateless configuration.

This server will associate the subnet with the client if one of the following conditions is met:

  • The DHCPINFORM is relayed and the giaddr matches the configured subnet.
  • The DHCPINFORM is unicast from the client and the ciaddr matches the configured subnet.
  • The DHCPINFORM is unicast from the client and the ciaddr is not set, but the source address of the IP packet matches the configured subnet.
  • The DHCPINFORM is not relayed and the IP address on the interface on which the message is received matches the configured subnet.

8.2.18. Client Classification in DHCPv4

The DHCPv4 server includes support for client classification. For a deeper discussion of the classification process, see Client Classification.

In certain cases it is useful to configure the server to differentiate between DHCP client types and treat them accordingly. Client classification can be used to modify the behavior of almost any part of DHCP message processing. Kea currently offers client classification via private options and option 43 deferred unpacking; subnet selection; pool selection; assignment of different options; and, for cable modems, specific options for use with the TFTP server address and the boot file field.

Kea can be instructed to limit access to given subnets based on class information. This is particularly useful for cases where two types of devices share the same link and are expected to be served from two different subnets. The primary use case for such a scenario is cable networks, where there are two classes of devices: the cable modem itself, which should be handed a lease from subnet A; and all other devices behind the modem, which should get leases from subnet B. That segregation is essential to prevent overly curious end-users from playing with their cable modems. For details on how to set up class restrictions on subnets, see Configuring Subnets With Class Information.

When subnets belong to a shared network, the classification applies to subnet selection but not to pools; that is, a pool in a subnet limited to a particular class can still be used by clients which do not belong to the class, if the pool they are expected to use is exhausted. The limit on access based on class information is also available at the pool level within a subnet: see Configuring Pools With Class Information. This is useful when segregating clients belonging to the same subnet into different address ranges.

In a similar way, a pool can be constrained to serve only known clients, i.e. clients which have a reservation, using the built-in KNOWN or UNKNOWN classes. Addresses can be assigned to registered clients without giving a different address per reservation: for instance, when there are not enough available addresses. The determination whether there is a reservation for a given client is made after a subnet is selected, so it is not possible to use KNOWN/UNKNOWN classes to select a shared network or a subnet.

The process of classification is conducted in five steps. The first step is to assess an incoming packet and assign it to zero or more classes. The second step is to choose a subnet, possibly based on the class information. When the incoming packet is in the special class DROP, it is dropped and a debug message logged. The next step is to evaluate class expressions depending on the built-in KNOWN/UNKNOWN classes after host reservation lookup, using them for pool selection and assigning classes from host reservations. The list of required classes is then built and each class of the list has its expression evaluated; when it returns true, the packet is added as a member of the class. The last step is to assign options, again possibly based on the class information. More complete and detailed information is available in Client Classification.

There are two main methods of classification. The first is automatic and relies on examining the values in the vendor class options or the existence of a host reservation. Information from these options is extracted, and a class name is constructed from it and added to the class list for the packet. The second method specifies an expression that is evaluated for each packet. If the result is true, the packet is a member of the class.

Note

The new early-global-reservations-lookup global parameter flag enables a lookup for global reservations before the subnet selection phase. This lookup is similar to the general lookup described above with two differences:

  • the lookup is limited to global host reservations
  • the UNKNOWN class is never set

Note

Care should be taken with client classification, as it is easy for clients that do not meet class criteria to be denied all service.

8.2.18.1. Setting Fixed Fields in Classification

It is possible to specify that clients belonging to a particular class should receive packets with specific values in certain fixed fields. In particular, three fixed fields are supported: next-server (conveys an IPv4 address, which is set in the siaddr field), server-hostname (conveys a server hostname, can be up to 64 bytes long, and is sent in the sname field) and boot-file-name (conveys the configuration file, can be up to 128 bytes long, and is sent using the file field).

Obviously, there are many ways to assign clients to specific classes, but for PXE clients the client architecture type option (code 93) seems to be particularly suited to make the distinction. The following example checks whether the client identifies itself as a PXE device with architecture EFI x86-64, and sets several fields if it does. See Section 2.1 of RFC 4578) or the client documentation for specific values.

"Dhcp4": {
    "client-classes": [
        {
            "name": "ipxe_efi_x64",
            "test": "option[93].hex == 0x0009",
            "next-server": "192.0.2.254",
            "server-hostname": "hal9000",
            "boot-file-name": "/dev/null"
        },
        ...
    ],
    ...
}

If an incoming packet is matched to multiple classes, then the value used for each field will come from the first class that specifies the field, in the order the classes are assigned to the packet.

Note

The classes are ordered as specified in the configuration.

8.2.18.2. Using Vendor Class Information in Classification

The server checks whether an incoming packet includes the vendor class identifier option (60). If it does, the content of that option is prepended with VENDOR_CLASS_, and it is interpreted as a class. For example, modern cable modems send this option with value docsis3.0, so the packet belongs to the class VENDOR_CLASS_docsis3.0.

Note

Certain special actions for clients in VENDOR_CLASS_docsis3.0 can be achieved by defining VENDOR_CLASS_docsis3.0 and setting its next-server and boot-file-name values appropriately.

This example shows a configuration using an automatically generated VENDOR_CLASS_ class. The administrator of the network has decided that addresses from the range 192.0.2.10 to 192.0.2.20 are going to be managed by the Dhcp4 server and only clients belonging to the DOCSIS 3.0 client class are allowed to use that pool.

"Dhcp4": {
    "subnet4": [
        {
            "id": 1,
            "subnet": "192.0.2.0/24",
            "pools": [ { "pool": "192.0.2.10 - 192.0.2.20" } ],
            "client-class": "VENDOR_CLASS_docsis3.0"
        }
    ],
    ...
}

8.2.18.3. Defining and Using Custom Classes

The following example shows how to configure a class using an expression and a subnet using that class. This configuration defines the class named Client_foo. It is comprised of all clients whose client IDs (option 61) start with the string foo. Members of this class will be given addresses from 192.0.2.10 to 192.0.2.20 and the addresses of their DNS servers set to 192.0.2.1 and 192.0.2.2.

"Dhcp4": {
    "client-classes": [
        {
            "name": "Client_foo",
            "test": "substring(option[61].hex,0,3) == 'foo'",
            "option-data": [
                {
                    "name": "domain-name-servers",
                    "code": 6,
                    "space": "dhcp4",
                    "csv-format": true,
                    "data": "192.0.2.1, 192.0.2.2"
                }
            ]
        },
        ...
    ],
    "subnet4": [
        {
            "id": 1,
            "subnet": "192.0.2.0/24",
            "pools": [ { "pool": "192.0.2.10 - 192.0.2.20" } ],
            "client-class": "Client_foo"
        },
        ...
    ],
    ...
}

8.2.18.4. Required Classification

In some cases it is useful to limit the scope of a class to a shared network, subnet, or pool. There are two parameters which are used to limit the scope of the class by instructing the server to evaluate test expressions when required.

The first one is the per-class only-if-required flag, which is false by default. When it is set to true, the test expression of the class is not evaluated at the reception of the incoming packet but later, and only if the class evaluation is required.

The second is require-client-classes, which takes a list of class names and is valid in shared-network, subnet, and pool scope. Classes in these lists are marked as required and evaluated after selection of this specific shared network/subnet/pool and before output-option processing.

In this example, a class is assigned to the incoming packet when the specified subnet is used:

"Dhcp4": {
    "client-classes": [
       {
           "name": "Client_foo",
           "test": "member('ALL')",
           "only-if-required": true
       },
       ...
    ],
    "subnet4": [
        {
            "subnet": "192.0.2.0/24",
            "pools": [ { "pool": "192.0.2.10 - 192.0.2.20" } ],
            "require-client-classes": [ "Client_foo" ],
            ...
        },
        ...
    ],
    ...
}

Required evaluation can be used to express complex dependencies like subnet membership. It can also be used to reverse the precedence; if option-data is set in a subnet, it takes precedence over option-data in a class. If option-data is moved to a required class and required in the subnet, a class evaluated earlier may take precedence.

Required evaluation is also available at the shared-network and pool levels. The order in which required classes are considered is: shared-network, subnet, and pool, i.e. in the reverse order from the way in which option-data is processed.

Note

Vendor-Identifying Vendor Options are a special case: for all other options an option is identified by its code point, but vivco-suboptions (124) and vivso-suboptions (125) are identified by the pair of code point and vendor identifier. This has no visible effect for vivso-suboptions, whose value is the vendor identifier, but it is different for vivco-suboptions, where the value is a record with the vendor identifier and a binary value. For instance, in:

"Dhcp4": {
    "option-data": [
       {
           "name": "vivco-suboptions",
           "always-send": true,
           "data": "1234, 03666f6f"
       },
       {
           "name": "vivco-suboptions",
           "always-send": true,
           "data": "5678, 03626172"
       },
       ...
     ],
     ...
 }

The first option-data entry does not hide the second one, because vendor identifiers (1234 and 5678) are different: the responses will carry two instances of the vivco-suboptions option, each for a different vendor.

8.2.19. DDNS for DHCPv4

As mentioned earlier, kea-dhcp4 can be configured to generate requests to the DHCP-DDNS server, kea-dhcp-ddns, (referred to herein as "D2") to update DNS entries. These requests are known as NameChangeRequests or NCRs. Each NCR contains the following information:

  1. Whether it is a request to add (update) or remove DNS entries.
  2. Whether the change requests forward DNS updates (A records), reverse DNS updates (PTR records), or both.
  3. The Fully Qualified Domain Name (FQDN), lease address, and DHCID (information identifying the client associated with the FQDN).

DDNS-related parameters are split into two groups:

  1. Connectivity Parameters

    These are parameters which specify where and how kea-dhcp4 connects to and communicates with D2. These parameters can only be specified within the top-level dhcp-ddns section in the kea-dhcp4 configuration. The connectivity parameters are listed below:

    • enable-updates
    • server-ip
    • server-port
    • sender-ip
    • sender-port
    • max-queue-size
    • ncr-protocol
    • ncr-format"
  2. Behavioral Parameters

    These parameters influence behavior such as how client host names and FQDN options are handled. They have been moved out of the dhcp-ddns section so that they may be specified at the global, shared-network, and/or subnet levels. Furthermore, they are inherited downward from global to shared-network to subnet. In other words, if a parameter is not specified at a given level, the value for that level comes from the level above it. The behavioral parameters are as follows:

    • ddns-send-updates
    • ddns-override-no-update
    • ddns-override-client-update
    • ddns-replace-client-name"
    • ddns-generated-prefix
    • ddns-qualifying-suffix
    • ddns-update-on-renew
    • ddns-conflict-resolution-mode
    • ddns-ttl-percent
    • hostname-char-set
    • hostname-char-replacement

Note

For backward compatibility, configuration parsing still recognizes the original behavioral parameters specified in dhcp-ddns, by translating the parameter into its global equivalent. If a parameter is specified both globally and in dhcp-ddns, the latter value is ignored. In either case, a log is emitted explaining what has occurred. Specifying these values within dhcp-ddns is deprecated and support for it will be removed.

The default configuration and values would appear as follows:

"Dhcp4": {
     "dhcp-ddns": {
        // Connectivity parameters
        "enable-updates": false,
         "server-ip": "127.0.0.1",
         "server-port":53001,
         "sender-ip":"",
         "sender-port":0,
         "max-queue-size":1024,
         "ncr-protocol":"UDP",
         "ncr-format":"JSON"
     },

     // Behavioral parameters (global)
     "ddns-send-updates": true,
     "ddns-override-no-update": false,
     "ddns-override-client-update": false,
     "ddns-replace-client-name": "never",
     "ddns-generated-prefix": "myhost",
     "ddns-qualifying-suffix": "",
     "ddns-update-on-renew": false,
     "ddns-conflict-resolution-mode": "check-with-dhcid",
     "hostname-char-set": "",
     "hostname-char-replacement": "",
     ...
}

There are two parameters which determine if kea-dhcp4 can generate DDNS requests to D2: the existing dhcp-ddns:enable-updates parameter, which now only controls whether kea-dhcp4 connects to D2; and the new behavioral parameter, ddns-send-updates, which determines whether DDNS updates are enabled at a given level (i.e. global, shared-network, or subnet). The following table shows how the two parameters function together:

Enabling and disabling DDNS updates
dhcp-ddns: enable-updates Global ddns-send-updates Outcome
false (default) false no updates at any scope
false true (default) no updates at any scope
true false updates only at scopes with a local value of true for ddns-enable-updates
true true updates at all scopes except those with a local value of false for ddns-enable-updates

Kea 1.9.1 added two new parameters; the first is ddns-update-on-renew. Normally, when leases are renewed, the server only updates DNS if the DNS information for the lease (e.g. FQDN, DNS update direction flags) has changed. Setting ddns-update-on-renew to true instructs the server to always update the DNS information when a lease is renewed, even if its DNS information has not changed. This allows Kea to "self-heal" if it was previously unable to add DNS entries or they were somehow lost by the DNS server.

Note

Setting ddns-update-on-renew to true may impact performance, especially for servers with numerous clients that renew often.

The second parameter added in Kea 1.9.1 is ddns-use-conflict-resolution. This boolean parameter was passed through to D2 and enabled or disabled conflict resolution as described in RFC 4703. Beginning with Kea 2.5.0, it is deprecated and replaced by ddns-conflict-resolution-mode which offers four modes of conflict resolution-related behavior:

  • check-with-dhcid - The default mode, it instructs D2 to carry out RFC 4703-compliant conflict resolution. Existing DNS entries may only be overwritten if they have a DHCID record and it matches the client's DHCID. This is equivalent to ddns-use-conflict-resolution: true;
  • no-check-with-dhcid - Existing DNS entries may be overwritten by any client, whether or not those entries include a DHCID record. The new entries will include a DHCID record for the client to whom they belong. This is equivalent to ddns-use-conflict-resolution: false;
  • check-exists-with-dhcid - Existing DNS entries may only be overwritten if they have a DHCID record. The DHCID record need not match the client's DHCID. This mode provides a way to protect static DNS entries (those that do not have a DHCID record) while allowing dynamic entries (those that do have a DHCID record) to be overwritten by any client. This behavior was not supported prior to Kea 2.4.0.
  • no-check-without-dhcid - Existing DNS entries may be overwritten by any client. New entries will not include DHCID records. This behavior was not supported prior to Kea 2.4.0.

Note

For backward compatibility, ddns-use-conflict-resolution is still accepted in JSON configuration. The server will replace the value internally, with the ddns-conflict-resolution-mode and an appropriate vavlue: ` check-with-dhcid` for true and no-check-with-dhcid for false.

Note

Setting ddns-conflict-resolution-mode to any value other than check-with-dhcid disables the one or more overwrite safeguards that the rules of conflict resolution (from RFC 4703) are intended to prevent. This means that existing entries for an FQDN or an IP address made for Client-A can be deleted or replaced by entries for Client-B. Furthermore, there are two scenarios by which entries for multiple clients for the same key (e.g. FQDN or IP) can be created.

1. Client-B uses the same FQDN as Client-A but a different IP address. In this case, the forward DNS entries (A and DHCID RRs) for Client-A will be deleted as they match the FQDN and new entries for Client-B will be added. The reverse DNS entries (PTR and DHCID RRs) for Client-A, however, will not be deleted as they belong to a different IP address, while new entries for Client-B will still be added.

2. Client-B uses the same IP address as Client-A but a different FQDN. In this case the reverse DNS entries (PTR and DHCID RRs) for Client-A will be deleted as they match the IP address, and new entries for Client-B will be added. The forward DNS entries (A and DHCID RRs) for Client-A, however, will not be deleted, as they belong to a different FQDN, while new entries for Client-B will still be added.

Disabling conflict resolution should be done only after careful review of specific use cases. The best way to avoid unwanted DNS entries is to always ensure lease changes are processed through Kea, whether they are released, expire, or are deleted via the lease4-del command, prior to reassigning either FQDNs or IP addresses. Doing so causes kea-dhcp4 to generate DNS removal requests to D2.

The DNS entries Kea creates contain a value for TTL (time to live). The kea-dhcp4 server calculates that value based on RFC 4702, Section 5, which suggests that the TTL value be 1/3 of the lease's lifetime, with a minimum value of 10 minutes.

The parameter ddns-ttl-percent, when specified, causes the TTL to be calculated as a simple percentage of the lease's lifetime, using the parameter's value as the percentage. It is specified as a decimal percent (e.g. .25, .75, 1.00) and may be specified at the global, shared-network, and subnet levels. By default it is unspecified.

8.2.19.1. DHCP-DDNS Server Connectivity

For NCRs to reach the D2 server, kea-dhcp4 must be able to communicate with it. kea-dhcp4 uses the following configuration parameters to control this communication:

  • enable-updates - Enables connectivity to kea-dhcp-ddns such that DDNS updates can be constructed and sent. It must be true for NCRs to be generated and sent to D2. It defaults to false.
  • server-ip - This is the IP address on which D2 listens for requests. The default is the local loopback interface at address 127.0.0.1. Either an IPv4 or IPv6 address may be specified.
  • server-port - This is the port on which D2 listens for requests. The default value is 53001.
  • sender-ip - This is the IP address which kea-dhcp4 uses to send requests to D2. The default value is blank, which instructs kea-dhcp4 to select a suitable address.
  • sender-port - This is the port which kea-dhcp4 uses to send requests to D2. The default value of 0 instructs kea-dhcp4 to select a suitable port.
  • max-queue-size - This is the maximum number of requests allowed to queue while waiting to be sent to D2. This value guards against requests accumulating uncontrollably if they are being generated faster than they can be delivered. If the number of requests queued for transmission reaches this value, DDNS updating is turned off until the queue backlog has been sufficiently reduced. The intent is to allow the kea-dhcp4 server to continue lease operations without running the risk that its memory usage grows without limit. The default value is 1024.
  • ncr-protocol - This specifies the socket protocol to use when sending requests to D2. Currently only UDP is supported.
  • ncr-format - This specifies the packet format to use when sending requests to D2. Currently only JSON format is supported.

By default, kea-dhcp-ddns is assumed to be running on the same machine as kea-dhcp4, and all of the default values mentioned above should be sufficient. If, however, D2 has been configured to listen on a different address or port, these values must be altered accordingly. For example, if D2 has been configured to listen on 192.168.1.10 port 900, the following configuration is required:

"Dhcp4": {
    "dhcp-ddns": {
        "server-ip": "192.168.1.10",
        "server-port": 900,
        ...
    },
    ...
}

8.2.19.2. When Does the kea-dhcp4 Server Generate a DDNS Request?

The kea-dhcp4 server follows the behavior prescribed for DHCP servers in RFC 4702. It is important to keep in mind that kea-dhcp4 makes the initial decision of when and what to update and forwards that information to D2 in the form of NCRs. Carrying out the actual DNS updates and dealing with such things as conflict resolution are within the purview of D2 itself (see The DHCP-DDNS Server). This section describes when kea-dhcp4 generates NCRs and the configuration parameters that can be used to influence this decision. It assumes that both the connectivity parameter enable-updates and the behavioral parameter ddns-send-updates, are true.

In general, kea-dhcp4 generates DDNS update requests when:

  1. A new lease is granted in response to a DHCPREQUEST;
  2. An existing lease is renewed but the FQDN associated with it has changed; or
  3. An existing lease is released in response to a DHCPRELEASE.

In the second case, lease renewal, two DDNS requests are issued: one request to remove entries for the previous FQDN, and a second request to add entries for the new FQDN. In the third case, a lease release - a single DDNS request - to remove its entries will be made.

As for the first case, the decisions involved when granting a new lease are more complex. When a new lease is granted, kea-dhcp4 generates a DDNS update request if the DHCPREQUEST contains either the FQDN option (code 81) or the Host Name option (code 12). If both are present, the server uses the FQDN option. By default, kea-dhcp4 respects the FQDN N and S flags specified by the client as shown in the following table:

Default FQDN flag behavior
Client Flags:N-S Client Intent Server Response Server Flags:N-S-O
0-0 Client wants to do forward updates, server should do reverse updates Server generates reverse-only request 1-0-0
0-1 Server should do both forward and reverse updates Server generates request to update both directions 0-1-0
1-0 Client wants no updates done Server does not generate a request 1-0-0

The first row in the table above represents "client delegation." Here the DHCP client states that it intends to do the forward DNS updates and the server should do the reverse updates. By default, kea-dhcp4 honors the client's wishes and generates a DDNS request to the D2 server to update only reverse DNS data. The parameter ddns-override-client-update can be used to instruct the server to override client delegation requests. When this parameter is true, kea-dhcp4 disregards requests for client delegation and generates a DDNS request to update both forward and reverse DNS data. In this case, the N-S-O flags in the server's response to the client will be 0-1-1 respectively.

(Note that the flag combination N=1, S=1 is prohibited according to RFC 4702. If such a combination is received from the client, the packet will be dropped by kea-dhcp4.)

To override client delegation, set the following values in the configuration file:

"Dhcp4": {
    "ddns-override-client-update": true,
    ...
}

The third row in the table above describes the case in which the client requests that no DNS updates be done. The parameter ddns-override-no-update can be used to instruct the server to disregard the client's wishes. When this parameter is true, kea-dhcp4 generates DDNS update requests to kea-dhcp-ddns even if the client requests that no updates be done. The N-S-O flags in the server's response to the client will be 0-1-1.

To override client delegation, issue the following commands:

"Dhcp4": {
    "ddns-override-no-update": true,
    ...
}

The kea-dhcp4 server always generates DDNS update requests if the client request only contains the Host Name option. In addition, it includes an FQDN option in the response to the client with the FQDN N-S-O flags set to 0-1-0, respectively. The domain name portion of the FQDN option is the name submitted to D2 in the DDNS update request.

8.2.19.3. kea-dhcp4 Name Generation for DDNS Update Requests

Each NameChangeRequest must of course include the fully qualified domain name whose DNS entries are to be affected. kea-dhcp4 can be configured to supply a portion or all of that name, based on what it receives from the client in the DHCPREQUEST.

The default rules for constructing the FQDN that will be used for DNS entries are:

  1. If the DHCPREQUEST contains the client FQDN option, take the candidate name from there; otherwise, take it from the Host Name option.
  2. If the candidate name is a partial (i.e. unqualified) name, then add a configurable suffix to the name and use the result as the FQDN.
  3. If the candidate name provided is empty, generate an FQDN using a configurable prefix and suffix.
  4. If the client provides neither option, then take no DNS action.

These rules can be amended by setting the ddns-replace-client-name parameter, which provides the following modes of behavior:

  • never - use the name the client sent. If the client sent no name, do not generate one. This is the default mode.
  • always - replace the name the client sent. If the client sent no name, generate one for the client.
  • when-present - replace the name the client sent. If the client sent no name, do not generate one.
  • when-not-present - use the name the client sent. If the client sent no name, generate one for the client.

Note

In early versions of Kea, this parameter was a boolean and permitted only values of true and false. Boolean values have been deprecated and are no longer accepted. Administrators currently using booleans must replace them with the desired mode name. A value of true maps to when-present, while false maps to never.

For example, to instruct kea-dhcp4 to always generate the FQDN for a client, set the parameter ddns-replace-client-name to always as follows:

"Dhcp4": {
    "ddns-replace-client-name": "always",
    ...
}

The prefix used in the generation of an FQDN is specified by the ddns-generated-prefix parameter. The default value is "myhost". To alter its value, simply set it to the desired string:

"Dhcp4": {
    "ddns-generated-prefix": "another.host",
    ...
}

The suffix used when generating an FQDN, or when qualifying a partial name, is specified by the ddns-qualifying-suffix parameter. It is strongly recommended that the user supply a value for the qualifying prefix when DDNS updates are enabled. For obvious reasons, we cannot supply a meaningful default.

"Dhcp4": {
    "ddns-qualifying-suffix": "foo.example.org",
    ...
}

When generating a name, kea-dhcp4 constructs the name in the format:

[ddns-generated-prefix]-[address-text].[ddns-qualifying-suffix].

where address-text is simply the lease IP address converted to a hyphenated string. For example, if the lease address is 172.16.1.10, the qualifying suffix is "example.com", and the default value is used for ddns-generated-prefix, the generated FQDN is:

myhost-172-16-1-10.example.com.

8.2.19.4. Sanitizing Client Host Name and FQDN Names

Some DHCP clients may provide values in the Host Name option (option code 12) or FQDN option (option code 81) that contain undesirable characters. It is possible to configure kea-dhcp4 to sanitize these values. The most typical use case is ensuring that only characters that are permitted by RFC 1035 be included: A-Z, a-z, 0-9, and "-". This may be accomplished with the following two parameters:

  • hostname-char-set - a regular expression describing the invalid character set. This can be any valid, regular expression using POSIX extended expression syntax. Embedded nulls (0x00) are always considered an invalid character to be replaced (or omitted). The default is "[^A-Za-z0-9.-]". This matches any character that is not a letter, digit, dot, hyphen, or null.
  • hostname-char-replacement - a string of zero or more characters with which to replace each invalid character in the host name. An empty string causes invalid characters to be OMITTED rather than replaced. The default is "".

The following configuration replaces anything other than a letter, digit, dot, or hyphen with the letter "x":

"Dhcp4": {
    "hostname-char-set": "[^A-Za-z0-9.-]",
    "hostname-char-replacement": "x",
    ...
}

Thus, a client-supplied value of "myhost-$[123.org" would become "myhost-xx123.org". Sanitizing is performed only on the portion of the name supplied by the client, and it is performed before applying a qualifying suffix (if one is defined and needed).

Note

Name sanitizing is meant to catch the more common cases of invalid characters through a relatively simple character-replacement scheme. It is difficult to devise a scheme that works well in all cases, for both Host Name and FQDN options. Administrators who find they have clients with odd corner cases of character combinations that cannot be readily handled with this mechanism should consider writing a hook that can carry out sufficiently complex logic to address their needs.

If clients include domain names in the Host Name option and the administrator wants these preserved, they need to make sure that the dot, ".", is considered a valid character by the hostname-char-set expression, such as this: "[^A-Za-z0-9.-]". This does not affect dots in FQDN Option values. When scrubbing FQDNs, dots are treated as delimiters and used to separate the option value into individual domain labels that are scrubbed and then re-assembled.

If clients are sending values that differ only by characters considered as invalid by the hostname-char-set, be aware that scrubbing them will yield identical values. In such cases, DDNS conflict rules will permit only one of them to register the name.

Finally, given the latitude clients have in the values they send, it is virtually impossible to guarantee that a combination of these two parameters will always yield a name that is valid for use in DNS. For example, using an empty value for hostname-char-replacement could yield an empty domain label within a name, if that label consists only of invalid characters.

Note

It is possible to specify hostname-char-set and/or hostname-char-replacement at the global scope. This allows host names to be sanitized without requiring a dhcp-ddns entry. When a hostname-char parameter is defined at both the global scope and in a dhcp-ddns entry, the second (local) value is used.

8.2.20. Next Server (siaddr)

In some cases, clients want to obtain configuration from a TFTP server. Although there is a dedicated option for it, some devices may use the siaddr field in the DHCPv4 packet for that purpose. That specific field can be configured using the next-server directive. It is possible to define it in the global scope or for a given subnet only. If both are defined, the subnet value takes precedence. The value in the subnet can be set to "0.0.0.0", which means that next-server should not be sent. It can also be set to an empty string, which is equivalent to it not being defined at all; that is, it uses the global value.

The server-hostname (which conveys a server hostname, can be up to 64 bytes long, and is in the sname field) and boot-file-name (which conveys the configuration file, can be up to 128 bytes long, and is sent using the file field) directives are handled the same way as next-server.

"Dhcp4": {
    "next-server": "192.0.2.123",
    "boot-file-name": "/dev/null",
    "subnet4": [
        {
            "next-server": "192.0.2.234",
            "server-hostname": "some-name.example.org",
            "boot-file-name": "bootfile.efi",
            ...
        }
    ],
    ...
}

8.2.21. Echoing Client-ID (RFC 6842)

The original DHCPv4 specification (RFC 2131) states that the DHCPv4 server must not send back client-id options when responding to clients. However, in some cases that results in confused clients that do not have a MAC address or client-id; see RFC 6842 for details. That behavior changed with the publication of RFC 6842, which updated RFC 2131. That update states that the server must send the client-id if the client sent it, and that is Kea's default behavior. However, in some cases older devices that do not support RFC 6842 may refuse to accept responses that include the client-id option. To enable backward compatibility, an optional configuration parameter has been introduced. To configure it, use the following configuration statement:

"Dhcp4": {
    "echo-client-id": false,
    ...
}

8.2.22. Using Client Identifier and Hardware Address

The DHCP server must be able to identify the client from which it receives the message and distinguish it from other clients. There are many reasons why this identification is required; the most important ones are:

  • When the client contacts the server to allocate a new lease, the server must store the client identification information in the lease database as a search key.
  • When the client tries to renew or release the existing lease, the server must be able to find the existing lease entry in the database for this client, using the client identification information as a search key.
  • Some configurations use static reservations for the IP addresses and other configuration information. The server's administrator uses client identification information to create these static assignments.
  • In dual-stack networks there is often a need to correlate the lease information stored in DHCPv4 and DHCPv6 servers for a particular host. Using common identification information by the DHCPv4 and DHCPv6 clients allows the network administrator to achieve this correlation and better administer the network. Beginning with release 2.1.2, Kea supports DHCPv6 DUIDs embedded within DHCPv4 Client Identifier options as described in RFC 4361.

DHCPv4 uses two distinct identifiers which are placed by the client in the queries sent to the server and copied by the server to its responses to the client: chaddr and client-identifier. The former was introduced as a part of the BOOTP specification and it is also used by DHCP to carry the hardware address of the interface used to send the query to the server (MAC address for the Ethernet). The latter is carried in the client-identifier option, introduced in RFC 2132.

RFC 2131 indicates that the server may use both of these identifiers to identify the client but the client identifier, if present, takes precedence over chaddr. One of the reasons for this is that the client identifier is independent from the hardware used by the client to communicate with the server. For example, if the client obtained the lease using one network card and then the network card is moved to another host, the server will wrongly identify this host as the one which obtained the lease. Moreover, RFC 4361 gives the recommendation to use a DUID (see RFC 8415, the DHCPv6 specification) carried as a client identifier when dual-stack networks are in use to provide consistent identification information for the client, regardless of the type of protocol it is using. Kea adheres to these specifications, and the client identifier by default takes precedence over the value carried in the chaddr field when the server searches, creates, updates, or removes the client's lease.

When the server receives a DHCPDISCOVER or DHCPREQUEST message from the client, it tries to find out if the client already has a lease in the database; if it does, the server hands out that lease rather than allocates a new one. Each lease in the lease database is associated with the client identifier and/or chaddr. The server first uses the client identifier (if present) to search for the lease; if one is found, the server treats this lease as belonging to the client, even if the current chaddr and the chaddr associated with the lease do not match. This facilitates the scenario when the network card on the client system has been replaced and thus the new MAC address appears in the messages sent by the DHCP client. If the server fails to find the lease using the client identifier, it performs another lookup using the chaddr. If this lookup returns no result, the client is considered to not have a lease and a new lease is created.

A common problem reported by network operators is that poor client implementations do not use stable client identifiers, instead generating a new client identifier each time the client connects to the network. Another well-known case is when the client changes its client identifier during the multi-stage boot process (PXE). In such cases, the MAC address of the client's interface remains stable, and using the chaddr field to identify the client guarantees that the particular system is considered to be the same client, even though its client identifier changes.

To address this problem, Kea includes a configuration option which enables client identification using chaddr only. This instructs the server to ignore the client identifier during lease lookups and allocations for a particular subnet. Consider the following simplified server configuration:

{
  "Dhcp4": {
    "match-client-id": true,
    "subnet4": [
    {
        "id": 1,
        "subnet": "192.0.10.0/24",
        "pools": [ { "pool": "192.0.2.23-192.0.2.87" } ],
        "match-client-id": false
    },
    {
        "id": 1,
        "subnet": "10.0.0.0/8",
        "pools": [ { "pool": "10.0.0.23-10.0.2.99" } ]
    }
    ]
  }
}

The match-client-id parameter is a boolean value which controls this behavior. The default value of true indicates that the server will use the client identifier for lease lookups and chaddr if the first lookup returns no results. false means that the server will only use the chaddr to search for the client's lease. Whether the DHCID for DNS updates is generated from the client identifier or chaddr is controlled through the same parameter.

The match-client-id parameter may appear both in the global configuration scope and/or under any subnet declaration. In the example shown above, the effective value of the match-client-id will be false for the subnet 192.0.10.0/24, because the subnet-specific setting of the parameter overrides the global value of the parameter. The effective value of the match-client-id for the subnet 10.0.0.0/8 will be set to true, because the subnet declaration lacks this parameter and the global setting is by default used for this subnet. In fact, the global entry for this parameter could be omitted in this case, because true is the default value.

It is important to understand what happens when the client obtains its lease for one setting of the match-client-id and then renews it when the setting has been changed. First, consider the case when the client obtains the lease and the match-client-id is set to true. The server stores the lease information, including the client identifier (if supplied) and chaddr, in the lease database. When the setting is changed and the client renews the lease, the server will determine that it should use the chaddr to search for the existing lease. If the client has not changed its MAC address, the server should successfully find the existing lease. The client identifier associated with the returned lease will be ignored and the client will be allowed to use this lease. When the lease is renewed only the chaddr will be recorded for this lease, according to the new server setting.

In the second case, the client has the lease with only a chaddr value recorded. When the match-client-id setting is changed to true, the server will first try to use the client identifier to find the existing client's lease. This will return no results because the client identifier was not recorded for this lease. The server will then use the chaddr and the lease will be found. If the lease appears to have no client identifier recorded, the server will assume that this lease belongs to the client and that it was created with the previous setting of the match-client-id. However, if the lease contains a client identifier which is different from the client identifier used by the client, the lease will be assumed to belong to another client and a new lease will be allocated.

For a more visual representation of how Kea recognizes the same client, check How Kea Recognizes the Same Client In Different DHCP Messages.

8.2.23. Authoritative DHCPv4 Server Behavior

The original DHCPv4 specification (RFC 2131) states that if a client requests an address in the INIT-REBOOT state of which the server has no knowledge, the server must remain silent, except if the server knows that the client has requested an IP address from the wrong network. By default, Kea follows the behavior of the ISC dhcpd daemon instead of the specification and also remains silent if the client requests an IP address from the wrong network, because configuration information about a given network segment is not known to be correct. Kea only rejects a client's DHCPREQUEST with a DHCPNAK message if it already has a lease for the client with a different IP address. Administrators can override this behavior through the boolean authoritative (false by default) setting.

In authoritative mode, authoritative set to true, Kea always rejects INIT-REBOOT requests from unknown clients with DHCPNAK messages. The authoritative setting can be specified in global, shared-network, and subnet configuration scope and is automatically inherited from the parent scope, if not specified. All subnets in a shared-network must have the same authoritative setting.

8.2.24. DHCPv4-over-DHCPv6: DHCPv4 Side

The support of DHCPv4-over-DHCPv6 transport is described in RFC 7341 and is implemented using cooperating DHCPv4 and DHCPv6 servers. This section is about the configuration of the DHCPv4 side (the DHCPv6 side is described in DHCPv4-over-DHCPv6: DHCPv6 Side).

Note

DHCPv4-over-DHCPv6 support is experimental and the details of the inter-process communication may change; for instance, the support of port relay (RFC 8357) introduced an incompatible change. Both the DHCPv4 and DHCPv6 sides should be running the same version of Kea.

The dhcp4o6-port global parameter specifies the first of the two consecutive ports of the UDP sockets used for the communication between the DHCPv6 and DHCPv4 servers. The DHCPv4 server is bound to ::1 on port + 1 and connected to ::1 on port.

With DHCPv4-over-DHCPv6, the DHCPv4 server does not have access to several of the identifiers it would normally use to select a subnet. To address this issue, three new configuration entries are available; the presence of any of these allows the subnet to be used with DHCPv4-over-DHCPv6. These entries are:

  • 4o6-subnet: takes a prefix (i.e., an IPv6 address followed by a slash and a prefix length) which is matched against the source address.
  • 4o6-interface-id: takes a relay interface ID option value.
  • 4o6-interface: takes an interface name which is matched against the incoming interface name.

ISC tested the following configuration:

{

# DHCPv4 conf
"Dhcp4": {

    "interfaces-config": {
        "interfaces": [ "eno33554984" ]
    },

    "lease-database": {
        "type": "memfile",
        "name": "leases4"
    },

    "valid-lifetime": 4000,

    "subnet4": [
    {
        "id": 1,
        "subnet": "10.10.10.0/24",
        "4o6-interface": "eno33554984",
        "4o6-subnet": "2001:db8:1:1::/64",
        "pools": [ { "pool": "10.10.10.100 - 10.10.10.199" } ]
    }
    ],

    "dhcp4o6-port": 6767,

    "loggers": [
    {
        "name": "kea-dhcp4",
        "output-options": [
        {
            "output": "/tmp/kea-dhcp4.log"
        }
        ],
        "severity": "DEBUG",
        "debuglevel": 0
    }
    ]
}

}

8.2.25. Sanity Checks in DHCPv4

An important aspect of a well-running DHCP system is an assurance that the data remains consistent; however, in some cases it may be convenient to tolerate certain inconsistent data. For example, a network administrator who temporarily removes a subnet from a configuration would not want all the leases associated with it to disappear from the lease database. Kea has a mechanism to implement sanity checks for situations like this.

Kea supports a configuration scope called sanity-checks. A parameter, called lease-checks, governs the verification carried out when a new lease is loaded from a lease file. This mechanism permits Kea to attempt to correct inconsistent data.

Every subnet has a subnet-id value; this is how Kea internally identifies subnets. Each lease has a subnet-id parameter as well, which identifies the subnet it belongs to. However, if the configuration has changed, it is possible that a lease could exist with a subnet-id but without any subnet that matches it. Also, it is possible that the subnet's configuration has changed and the subnet-id now belongs to a subnet that does not match the lease.

Kea's corrective algorithm first checks to see if there is a subnet with the subnet-id specified by the lease. If there is, it verifies whether the lease belongs to that subnet. If not, depending on the lease-checks setting, the lease is discarded, a warning is displayed, or a new subnet is selected for the lease that matches it topologically.

There are five levels which are supported:

  • none - do no special checks; accept the lease as is.
  • warn - if problems are detected display a warning, but accept the lease data anyway. This is the default value.
  • fix - if a data inconsistency is discovered, try to correct it. If the correction is not successful, insert the incorrect data anyway.
  • fix-del - if a data inconsistency is discovered, try to correct it. If the correction is not successful, reject the lease. This setting ensures the data's correctness, but some incorrect data may be lost. Use with care.
  • del - if any inconsistency is detected, reject the lease. This is the strictest mode; use with care.

This feature is currently implemented for the memfile backend. The sanity check applies to the lease database in memory, not to the lease file, i.e. inconsistent leases will stay in the lease file.

An example configuration that sets this parameter looks as follows:

"Dhcp4": {
    "sanity-checks": {
        "lease-checks": "fix-del"
    },
    ...
}

8.2.26. Storing Extended Lease Information

To support such features as DHCP Leasequery (RFC 4388), additional information must be stored with each lease. Because the amount of information for each lease has ramifications in terms of performance and system resource consumption, storage of this additional information is configurable through the store-extended-info parameter. It defaults to false and may be set at the global, shared-network, and subnet levels.

"Dhcp4": {
    "store-extended-info": true,
    ...
}

When set to true, information relevant to the DHCPREQUEST asking for the lease is added into the lease's user-context as a map element labeled "ISC". Since Kea version 2.3.2, when the DHCPREQUEST received contains the option (DHCP Option 82), the map contains the relay-agent-info map with the content option (DHCP Option 82) in the sub-options entry and, when present, the remote-id and relay-id options. Since DHCPREQUESTs sent as renewals are not likely to contain this information, the values taken from the last DHCPREQUEST that did contain it are retained on the lease. The lease's user-context looks something like this:

{ "ISC": { "relay-agent-info": { "sub-options": "0x0104AABBCCDD" } } }

Or with remote and relay sub-options:

{
    "ISC": {
        "relay-agent-info": {
            "sub-options": "0x02030102030C03AABBCC",
            "remote-id": "03010203",
            "relay-id": "AABBCC"
        }
    }
}

Note

It is possible that other hook libraries are already using user-context. Enabling store-extended-info should not interfere with any other user-context content, as long as it does not also use an element labeled "ISC". In other words, user-context is intended to be a flexible container serving multiple purposes. As long as no other purpose also writes an "ISC" element to user-context there should not be a conflict.

Extended lease information is also subject to configurable sanity checking. The parameter in the sanity-checks scope is named extended-info-checks and supports these levels:

  • none - do no check nor upgrade. This level should be used only when extended info is not used at all or when no badly formatted extended info, including using the old format, is expected.
  • fix - fix some common inconsistencies and upgrade extended info using the old format to the new one. It is the default level and is convenient when the Leasequery hook library is not loaded.
  • strict - fix all inconsistencies which have an impact on the (Bulk) Leasequery hook library.
  • pedantic - enforce full conformance to the format produced by the Kea code; for instance, no extra entries are allowed with the exception of comment.

Note

This feature is currently implemented only for the memfile backend. The sanity check applies to the lease database in memory, not to the lease file, i.e. inconsistent leases stay in the lease file.

8.2.27. Multi-Threading Settings

The Kea server can be configured to process packets in parallel using multiple threads. These settings can be found under the multi-threading structure and are represented by:

  • enable-multi-threading - use multiple threads to process packets in parallel. The default is true.
  • thread-pool-size - specify the number of threads to process packets in parallel. It may be set to 0 (auto-detect), or any positive number that explicitly sets the thread count. The default is 0.
  • packet-queue-size - specify the size of the queue used by the thread pool to process packets. It may be set to 0 (unlimited), or any positive number that explicitly sets the queue size. The default is 64.

An example configuration that sets these parameters looks as follows:

"Dhcp4": {
    "multi-threading": {
       "enable-multi-threading": true,
       "thread-pool-size": 4,
       "packet-queue-size": 16
    },
    ...
}

8.2.28. Multi-Threading Settings With Different Database Backends

The Kea DHCPv4 server is benchmarked by ISC to determine which settings give the best performance. Although this section describes our results, they are merely recommendations and are very dependent on the particular hardware used for benchmarking. We strongly advise that administrators run their own performance benchmarks.

A full report of performance results for the latest stable Kea version can be found here. This includes hardware and benchmark scenario descriptions, as well as current results.

After enabling multi-threading, the number of threads is set by the thread-pool-size parameter. Results from our experiments show that the best settings for kea-dhcp4 are:

  • thread-pool-size: 4 when using memfile for storing leases.
  • thread-pool-size: 12 or more when using mysql for storing leases.
  • thread-pool-size: 8 when using postgresql.

Another very important parameter is packet-queue-size; in our benchmarks we used it as a multiplier of thread-pool-size. The actual setting strongly depends on thread-pool-size.

We saw the best results in our benchmarks with the following settings:

  • packet-queue-size: 7 * thread-pool-size when using memfile for storing leases; in our case it was 7 * 4 = 28. This means that at any given time, up to 28 packets could be queued.
  • packet-queue-size: 66 * thread-pool-size when using mysql for storing leases; in our case it was 66 * 12 = 792. This means that up to 792 packets could be queued.
  • packet-queue-size: 11 * thread-pool-size when using postgresql for storing leases; in our case it was 11 * 8 = 88.

8.2.29. IPv6-Only Preferred Networks

RFC8925, recently published by the IETF, specifies a DHCPv4 option to indicate that a host supports an IPv6-only mode and is willing to forgo obtaining an IPv4 address if the network provides IPv6 connectivity. The general idea is that a network administrator can enable this option to signal to compatible dual-stack devices that IPv6 connectivity is available and they can shut down their IPv4 stack. The new option v6-only-preferred content is a 32-bit unsigned integer and specifies for how long the device should disable its stack. The value is expressed in seconds.

The RFC mentions the V6ONLY_WAIT timer. This is implemented in Kea by setting the value of the v6-only-preferred option. This follows the usual practice of setting options; the option value can be specified on the pool, subnet, shared network, or global levels, or even via host reservations.

There is no special processing involved; it follows the standard Kea option processing regime. The option is not sent back unless the client explicitly requests it. For example, to enable the option for the whole subnet, the following configuration can be used:

{
"subnet4": [
    {
        "id": 1,
        "pools": [ { "pool":  "192.0.2.1 - 192.0.2.200" } ],
        "subnet": "192.0.2.0/24",
        "option-data": [
            {
                // This will make the v6-only capable devices to disable their
                // v4 stack for half an hour and then try again
                "name": "v6-only-preferred",
                "data": "1800"
            }
        ]
    }
],
...
}

8.2.30. Lease Caching

Clients that attempt multiple renewals in a short period can cause the server to update and write to the database frequently, resulting in a performance impact on the server. The cache parameters instruct the DHCP server to avoid updating leases too frequently, thus avoiding this behavior. Instead, the server assigns the same lease (i.e. reuses it) with no modifications except for CLTT (Client Last Transmission Time), which does not require disk operations.

The two parameters are the cache-threshold double and the cache-max-age integer; they have no default setting, i.e. the lease caching feature must be explicitly enabled. These parameters can be configured at the global, shared-network, and subnet levels. The subnet level has precedence over the shared-network level, while the global level is used as a last resort. For example:

{
"subnet4": [
    {
        "pools": [ { "pool":  "192.0.2.1 - 192.0.2.200" } ],
        "subnet": "192.0.2.0/24",
        "cache-threshold": .25,
        "cache-max-age": 600,
        "valid-lifetime": 2000,
        ...
    }
],
...
}

When an already-assigned lease can fulfill a client query:

  • any important change, e.g. for DDNS parameter, hostname, or valid lifetime reduction, makes the lease not reusable.
  • lease age, i.e. the difference between the creation or last modification time and the current time, is computed (elapsed duration).
  • if cache-max-age is explicitly configured, it is compared with the lease age; leases that are too old are not reusable. This means that the value 0 for cache-max-age disables the lease cache feature.
  • if cache-threshold is explicitly configured and is between 0.0 and 1.0, it expresses the percentage of the lease valid lifetime which is allowed for the lease age. Values below and including 0.0 and values greater than 1.0 disable the lease cache feature.

In our example, a lease with a valid lifetime of 2000 seconds can be reused if it was committed less than 500 seconds ago. With a lifetime of 3000 seconds, a maximum age of 600 seconds applies.

In outbound client responses (e.g. DHCPACK messages), the dhcp-lease-time option is set to the reusable valid lifetime, i.e. the expiration date does not change. Other options based on the valid lifetime e.g. dhcp-renewal-time and dhcp-rebinding-time, also depend on the reusable lifetime.

8.2.31. Temporary Allocation on DHCPDISCOVER

By default, kea-dhcp4 does not allocate or store a lease when offering an address to a client in response to a DHCPDISCOVER. In general, kea-dhcp4 can fulfill client demands faster by deferring lease allocation and storage until it receives DHCPREQUESTs for them. Release 2.3.6 added a new parameter to kea-dhcp4, offer-lifetime, which (when not zero) instructs the server to allocate and persist a lease when generating a DHCPOFFER. In addition:

  • The persisted lease's lifetime is equal to offer-lifetime (in seconds).
  • The lifetime sent to the client in the DHCPOFFER via option 51 is still based on valid-lifetime. This avoids issues with clients that may reject offers whose lifetimes they perceive as too short.
  • DDNS updates are not performed. As with the default behavior, those updates occur on DHCPREQUEST.
  • Updates are not sent to HA peers.
  • Assigned lease statistics are incremented.
  • Expiration processing and reclamation behave just as they do for leases allocated during DHCPREQUEST processing.
  • Lease caching, if enabled, is honored.
  • In sites running multiple instances of kea-dhcp4 against a single, shared lease store, races for given address values are lost during DHCPDISCOVER processing rather than during DHCPREQUEST processing. Servers that lose the race for the address simply do not respond to the client, rather than NAK them. The client in turn simply retries its DHCPDISCOVER. This should reduce the amount of traffic such conflicts incur.
  • Clients repeating DHCPDISCOVERs are offered the same address each time.

An example subnet configuration is shown below:

{
"subnet4": [
    {
        "pools": [ { "pool":  "192.0.2.1 - 192.0.2.200" } ],
        "subnet": "192.0.2.0/24",
        "offer-lifetime": 60,
        "valid-lifetime": 2000,
        ...
    }
],
...
}

Here offer-lifetime has been configured to be 60 seconds, with a valid-lifetime of 2000 seconds. This instructs kea-dhcp4 to persist leases for 60 seconds when sending them back in DHCPOFFERs, and then extend them to 2000 seconds when clients DHCPREQUEST them.

The value, which defaults to 0, is supported at the global, shared-network, subnet, and class levels. Choosing an appropriate value for offer-lifetime is extremely site-dependent, but a value between 60 and 120 seconds is a reasonable starting point.

8.3. Host Reservations in DHCPv4

There are many cases where it is useful to provide a configuration on a per-host basis. The most obvious one is to reserve a specific, static address for exclusive use by a given client (host); the returning client receives the same address from the server every time, and other clients generally do not receive that address. Host reservations are also convenient when a host has specific requirements, e.g. a printer that needs additional DHCP options. Yet another possible use case is to define unique names for hosts.

There may be cases when a new reservation has been made for a client for an address currently in use by another client. We call this situation a "conflict." These conflicts get resolved automatically over time, as described in subsequent sections. Once a conflict is resolved, the correct client will receive the reserved configuration when it renews.

Host reservations are defined as parameters for each subnet. Each host must have its own unique identifier, such as the hardware/MAC address. There is an optional reservations array in the subnet4 structure; each element in that array is a structure that holds information about reservations for a single host. In particular, the structure has an identifier that uniquely identifies a host. In the DHCPv4 context, the identifier is usually a hardware or MAC address. In most cases an IP address will be specified. It is also possible to specify a hostname, host-specific options, or fields carried within the DHCPv4 message such as siaddr, sname, or file.

Note

The reserved address must be within the subnet.

The following example shows how to reserve addresses for specific hosts in a subnet:

{
"subnet4": [
    {
        "id": 1,
        "pools": [ { "pool":  "192.0.2.1 - 192.0.2.200" } ],
        "subnet": "192.0.2.0/24",
        "interface": "eth0",
        "reservations": [
            {
                "hw-address": "1a:1b:1c:1d:1e:1f",
                "ip-address": "192.0.2.202"
            },
            {
                "duid": "0a:0b:0c:0d:0e:0f",
                "ip-address": "192.0.2.100",
                "hostname": "alice-laptop"
            },
            {
                "circuit-id": "'charter950'",
                "ip-address": "192.0.2.203"
            },
            {
                "client-id": "01:11:22:33:44:55:66",
                "ip-address": "192.0.2.204"
            }
        ]
    }
],
...
}

The first entry reserves the 192.0.2.202 address for the client that uses a MAC address of 1a:1b:1c:1d:1e:1f. The second entry reserves the address 192.0.2.100 and the hostname of "alice-laptop" for the client using a DUID 0a:0b:0c:0d:0e:0f. (If DNS updates are planned, it is strongly recommended that the hostnames be unique.) The third example reserves address 192.0.3.203 for a client whose request would be relayed by a relay agent that inserts a circuit-id option with the value "charter950". The fourth entry reserves address 192.0.2.204 for a client that uses a client identifier with value 01:11:22:33:44:55:66.

The above example is used for illustrational purposes only; in actual deployments it is recommended to use as few types as possible (preferably just one). See Fine-Tuning DHCPv4 Host Reservation for a detailed discussion of this point.

Making a reservation for a mobile host that may visit multiple subnets requires a separate host definition in each subnet that host is expected to visit. It is not possible to define multiple host definitions with the same hardware address in a single subnet. Multiple host definitions with the same hardware address are valid if each is in a different subnet.

Adding host reservations incurs a performance penalty. In principle, when a server that does not support host reservation responds to a query, it needs to check whether there is a lease for a given address being considered for allocation or renewal. The server that does support host reservation has to perform additional checks: not only whether the address is currently used (i.e., if there is a lease for it), but also whether the address could be used by someone else (i.e., if there is a reservation for it). That additional check incurs extra overhead.

8.3.1. Address Reservation Types

In a typical Kea scenario there is an IPv4 subnet defined, e.g. 192.0.2.0/24, with a certain part of it dedicated for dynamic allocation by the DHCPv4 server. That dynamic part is referred to as a dynamic pool or simply a pool. In principle, a host reservation can reserve any address that belongs to the subnet. The reservations that specify addresses that belong to configured pools are called "in-pool reservations." In contrast, those that do not belong to dynamic pools are called "out-of-pool reservations." There is no formal difference in the reservation syntax and both reservation types are handled uniformly.

Kea supports global host reservations. These are reservations that are specified at the global level within the configuration and that do not belong to any specific subnet. Kea still matches inbound client packets to a subnet as before, but when the subnet's reservation mode is set to "global", Kea looks for host reservations only among the global reservations defined. Typically, such reservations would be used to reserve hostnames for clients which may move from one subnet to another.

Note

Global reservations, while useful in certain circumstances, have aspects that must be given due consideration when using them. Please see Conflicts in DHCPv4 Reservations for more details.

Note

Since Kea 1.9.1, reservation mode has been replaced by three boolean flags, reservations-global, reservations-in-subnet, and reservations-out-of-pool, which allow the configuration of host reservations both globally and in a subnet. In such cases a subnet host reservation has preference over a global reservation when both exist for the same client.

8.3.2. Conflicts in DHCPv4 Reservations

As reservations and lease information are stored separately, conflicts may arise. Consider the following series of events: the server has configured the dynamic pool of addresses from the range of 192.0.2.10 to 192.0.2.20. Host A requests an address and gets 192.0.2.10. Now the system administrator decides to reserve address 192.0.2.10 for Host B. In general, reserving an address that is currently assigned to someone else is not recommended, but there are valid use cases where such an operation is warranted.

The server now has a conflict to resolve. If Host B boots up and requests an address, the server cannot immediately assign the reserved address 192.0.2.10. A naive approach would to be immediately remove the existing lease for Host A and create a new one for Host B. That would not solve the problem, though, because as soon as Host B gets the address, it will detect that the address is already in use (by Host A) and will send a DHCPDECLINE message. Therefore, in this situation, the server has to temporarily assign a different address from the dynamic pool (not matching what has been reserved) to Host B.

When Host A renews its address, the server will discover that the address being renewed is now reserved for another host - Host B. The server will inform Host A that it is no longer allowed to use it by sending a DHCPNAK message. The server will not remove the lease, though, as there's a small chance that the DHCPNAK will not be delivered if the network is lossy. If that happens, the client will not receive any responses, so it will retransmit its DHCPREQUEST packet. Once the DHCPNAK is received by Host A, it will revert to server discovery and will eventually get a different address. Besides allocating a new lease, the server will also remove the old one. As a result, address 192.0.2.10 will become free.

When Host B tries to renew its temporarily assigned address, the server will detect that it has a valid lease, but will note that there is a reservation for a different address. The server will send DHCPNAK to inform Host B that its address is no longer usable, but will keep its lease (again, the DHCPNAK may be lost, so the server will keep it until the client returns for a new address). Host B will revert to the server discovery phase and will eventually send a DHCPREQUEST message. This time the server will find that there is a reservation for that host and that the reserved address 192.0.2.10 is not used, so it will be granted. It will also remove the lease for the temporarily assigned address that Host B previously obtained.

This recovery will succeed, even if other hosts attempt to get the reserved address. If Host C requests the address 192.0.2.10 after the reservation is made, the server will either offer a different address (when responding to DHCPDISCOVER) or send DHCPNAK (when responding to DHCPREQUEST).

This mechanism allows the server to fully recover from a case where reservations conflict with existing leases; however, this procedure takes roughly as long as the value set for renew-timer. The best way to avoid such a recovery is not to define new reservations that conflict with existing leases. Another recommendation is to use out-of-pool reservations; if the reserved address does not belong to a pool, there is no way that other clients can get it.

Note

The conflict-resolution mechanism does not work for global reservations. Although the global address reservations feature may be useful in certain settings, it is generally recommended not to use global reservations for addresses. Administrators who do choose to use global reservations must manually ensure that the reserved addresses are not in dynamic pools.

8.3.3. Reserving a Hostname

When the reservation for a client includes the hostname, the server returns this hostname to the client in the Client FQDN or Hostname option. The server responds with the Client FQDN option only if the client has included the Client FQDN option in its message to the server. The server responds with the Hostname option if the client included the Hostname option in its message to the server, or if the client requested the Hostname option using the Parameter Request List option. The server returns the Hostname option even if it is not configured to perform DNS updates. The reserved hostname always takes precedence over the hostname supplied by the client or the autogenerated (from the IPv4 address) hostname.

The server qualifies the reserved hostname with the value of the ddns-qualifying-suffix parameter. For example, the following subnet configuration:

{
    "subnet4": [
    {
        "id": 1,
        "subnet": "10.0.0.0/24",
        "pools": [ { "pool": "10.0.0.10-10.0.0.100" } ],
        "ddns-qualifying-suffix": "example.isc.org.",
        "reservations": [
           {
             "hw-address": "aa:bb:cc:dd:ee:ff",
             "hostname": "alice-laptop"
           }
        ]
    }
    ],
    "dhcp-ddns": {
        "enable-updates": true
    }
}

will result in the "alice-laptop.example.isc.org." hostname being assigned to the client using the MAC address "aa:bb:cc:dd:ee:ff". If the ddns-qualifying-suffix is not specified, the default (empty) value will be used, and in this case the value specified as a hostname will be treated as a fully qualified name. Thus, by leaving the ddns-qualifying-suffix empty it is possible to qualify hostnames for different clients with different domain names:

{
    "subnet4": [
    {
        "id": 1,
        "subnet": "10.0.0.0/24",
        "pools": [ { "pool": "10.0.0.10-10.0.0.100" } ],
        "reservations": [
           {
             "hw-address": "aa:bb:cc:dd:ee:ff",
             "hostname": "alice-laptop.isc.org."
           },
           {
             "hw-address": "12:34:56:78:99:AA",
             "hostname": "mark-desktop.example.org."
           }
        ]
    }
    ],
    "dhcp-ddns": {
        "enable-updates": true
    }
}

The above example results in the assignment of the "alice-laptop.isc.org." hostname to the client using the MAC address "aa:bb:cc:dd:ee:ff", and the hostname "mark-desktop.example.org." to the client using the MAC address "12:34:56:78:99:AA".

8.3.4. Including Specific DHCPv4 Options in Reservations

Kea offers the ability to specify options on a per-host basis. These options follow the same rules as any other options. These can be standard options (see Standard DHCPv4 Options), custom options (see Custom DHCPv4 Options), or vendor-specific options (see DHCPv4 Vendor-Specific Options). The following example demonstrates how standard options can be defined:

{
    "subnet4": [
    {
        "reservations": [
        {
            "hw-address": "aa:bb:cc:dd:ee:ff",
            "ip-address": "192.0.2.1",
            "option-data": [
            {
                "name": "cookie-servers",
                "data": "10.1.1.202,10.1.1.203"
            },
            {
                "name": "log-servers",
                "data": "10.1.1.200,10.1.1.201"
            }
            ]
        }
        ],
        ...
    }
    ],
    ...
}

Vendor-specific options can be reserved in a similar manner:

{
    "subnet4": [
    {
        "reservations": [
        {
            "hw-address": "aa:bb:cc:dd:ee:ff",
            "ip-address": "10.0.0.7",
            "option-data": [
            {
                "name": "vivso-suboptions",
                "data": "4491"
            },
            {
                "name": "tftp-servers",
                "space": "vendor-4491",
                "data": "10.1.1.202,10.1.1.203"
            }
            ]
        }
        ],
        ...
    }
    ],
    ...
}

Options defined at the host level have the highest priority. In other words, if there are options defined with the same type on the global, subnet, class, and host levels, the host-specific values are used.

8.3.5. Reserving Next Server, Server Hostname, and Boot File Name

BOOTP/DHCPv4 messages include "siaddr", "sname", and "file" fields. Even though DHCPv4 includes corresponding options, such as option 66 and option 67, some clients may not support these options. For this reason, server administrators often use the "siaddr", "sname", and "file" fields instead.

With Kea, it is possible to make static reservations for these DHCPv4 message fields:

{
    "subnet4": [
    {
        "reservations": [
        {
            "hw-address": "aa:bb:cc:dd:ee:ff",
            "next-server": "10.1.1.2",
            "server-hostname": "server-hostname.example.org",
            "boot-file-name": "/tmp/bootfile.efi"
        }
        ],
        ...
    }
    ],
    ...
}

Note that those parameters can be specified in combination with other parameters for a reservation, such as a reserved IPv4 address. These parameters are optional; a subset of them can be specified, or all of them can be omitted.

8.3.6. Reserving Client Classes in DHCPv4

Using Expressions in Classification explains how to configure the server to assign classes to a client, based on the content of the options that this client sends to the server. Host reservation mechanisms also allow for the static assignment of classes to clients. The definitions of these classes are placed in the Kea configuration file or a database. The following configuration snippet shows how to specify that a client belongs to the classes reserved-class1 and reserved-class2. Those classes are associated with specific options sent to the clients which belong to them.

{
    "client-classes": [
    {
       "name": "reserved-class1",
       "option-data": [
       {
           "name": "routers",
           "data": "10.0.0.200"
       }
       ]
    },
    {
       "name": "reserved-class2",
       "option-data": [
       {
           "name": "domain-name-servers",
           "data": "10.0.0.201"
       }
       ]
    }
    ],
    "subnet4": [
    {
        "id": 1,
        "subnet": "10.0.0.0/24",
        "pools": [ { "pool": "10.0.0.10-10.0.0.100" } ],
        "reservations": [
        {
            "hw-address": "aa:bb:cc:dd:ee:ff",

            "client-classes": [ "reserved-class1", "reserved-class2" ]

        }
        ]
    }
    ]
 }

In some cases the host reservations can be used in conjunction with client classes specified within the Kea configuration. In particular, when a host reservation exists for a client within a given subnet, the "KNOWN" built-in class is assigned to the client. Conversely, when there is no static assignment for the client, the "UNKNOWN" class is assigned to the client. Class expressions within the Kea configuration file can refer to "KNOWN" or "UNKNOWN" classes using the "member" operator. For example:

{
    "client-classes": [
        {
            "name": "dependent-class",
            "test": "member('KNOWN')",
            "only-if-required": true
        }
    ]
}

The only-if-required parameter is needed here to force evaluation of the class after the lease has been allocated and thus the reserved class has been also assigned.

Note

The classes specified in non-global host reservations are assigned to the processed packet after all classes with the only-if-required parameter set to false have been evaluated. This means that these classes must not depend on the statically assigned classes from the host reservations. If such a dependency is needed, the only-if-required parameter must be set to true for the dependent classes. Such classes are evaluated after the static classes have been assigned to the packet. This, however, imposes additional configuration overhead, because all classes marked as only-if-required must be listed in the require-client-classes list for every subnet where they are used.

Note

Client classes specified within the Kea configuration file may depend on the classes specified within the global host reservations. In such a case the only-if-required parameter is not needed. Refer to Pool Selection with Client Class Reservations and Subnet Selection with Client Class Reservations for specific use cases.

8.3.7. Storing Host Reservations in MySQL or PostgreSQL

Kea can store host reservations in MySQL or PostgreSQL. See Hosts Storage for information on how to configure Kea to use reservations stored in MySQL or PostgreSQL. Kea provides a dedicated hook for managing reservations in a database; section libdhcp_host_cmds.so: Host Commands provides detailed information. The Kea wiki provides some examples of how to conduct common host reservation operations.

Note

In Kea, the maximum length of an option specified per-host-reservation is arbitrarily set to 4096 bytes.

8.3.8. Fine-Tuning DHCPv4 Host Reservation

The host reservation capability introduces additional restrictions for the allocation engine (the component of Kea that selects an address for a client) during lease selection and renewal. In particular, three major checks are necessary. First, when selecting a new lease, it is not sufficient for a candidate lease to simply not be in use by another DHCP client; it also must not be reserved for another client. Similarly, when renewing a lease, an additional check must be performed to see whether the address being renewed is reserved for another client. Finally, when a host renews an address, the server must check whether there is a reservation for this host, which would mean the existing (dynamically allocated) address should be revoked and the reserved one be used instead.

Some of those checks may be unnecessary in certain deployments, and not performing them may improve performance. The Kea server provides the reservation-mode configuration parameter to select the types of reservations allowed for a particular subnet. Each reservation type has different constraints for the checks to be performed by the server when allocating or renewing a lease for the client. Although reservation-mode was deprecated in Kea 1.9.1, it is still available; the allowed values are:

  • all - enables both in-pool and out-of-pool host reservation types. This setting is the default value, and is the safest and most flexible. However, as all checks are conducted, it is also the slowest. It does not check against global reservations.
  • out-of-pool - allows only out-of-pool host reservations. With this setting in place, the server assumes that all host reservations are for addresses that do not belong to the dynamic pool. Therefore, it can skip the reservation checks when dealing with in-pool addresses, thus improving performance. Do not use this mode if any reservations use in-pool addresses. Caution is advised when using this setting; Kea does not sanity-check the reservations against reservation-mode and misconfiguration may cause problems.
  • global - allows only global host reservations. With this setting in place, the server searches for reservations for a client only among the defined global reservations. If an address is specified, the server skips the reservation checks carried out in other modes, thus improving performance. Caution is advised when using this setting; Kea does not sanity-check reservations when global is set, and misconfiguration may cause problems.
  • disabled - host reservation support is disabled. As there are no reservations, the server skips all checks. Any reservations defined are completely ignored. As checks are skipped, the server may operate faster in this mode.

Since Kea 1.9.1, the reservation-mode parameter is replaced by the reservations-global, reservations-in-subnet, and reservations-out-of-pool flags. The flags can be activated independently and can produce various combinations, some of which were not supported by the deprecated reservation-mode.

The reservation-mode parameter can be specified at:

  • global level: .Dhcp4["reservation-mode"] (lowest priority: gets overridden by all others)
  • subnet level: .Dhcp4.subnet4[]["reservation-mode"] (low priority)
  • shared-network level: .Dhcp4["shared-networks"][]["reservation-mode"] (high priority)
  • shared-network subnet-level: .Dhcp4["shared-networks"][].subnet4[]["reservation-mode"] (highest priority: overrides all others)

To decide which reservation-mode to choose, the following decision diagram may be useful:

                              O
                              |
                              v
+-----------------------------+------------------------------+
|         Is per-host configuration needed, such as          |
|                reserving specific addresses,               |
|               assigning specific options or                |
| assigning packets to specific classes on per-device basis? |
+-+-----------------+----------------------------------------+
  |                 |
no|              yes|
  |                 |   +--------------------------------------+
  |                 |   |         For all given hosts,         |
  +--> "disabled"   +-->+      can the reserved resources      |
                        |  be used in all configured subnets?  |
                        +--------+---------------------------+-+
                                 |                           |
+----------------------------+   |no                         |yes
|             Is             |   |                           |
|  at least one reservation  +<--+               "global" <--+
| used to reserve addresses? |
+-+------------------------+-+
  |                        |
no|                     yes|   +---------------------------+
  |                        |   | Is high leases-per-second |
  +--> "out-of-pool"       +-->+ performance or efficient  |
        ^                      |      resource usage       |
        |                      |  (CPU ticks, RAM usage,   |
        |                      |   database roundtrips)    |
        |                      | important to your setup?  |
        |                      +-+----------------+--------+
        |                        |                |
        |                     yes|              no|
        |                        |                |
        |          +-------------+                |
        |          |                              |
        |          |   +----------------------+   |
        |          |   | Can it be guaranteed |   |
        |          +-->+  that the reserved   |   |
        |              |      addresses       |   |
        |              |  aren't part of the  |   |
        |              |   pools configured   |   |
        |              |  in the respective   |   |
        |              |       subnet?        |   |
        |              +-+------------------+-+   |
        |                |                  |     |
        |             yes|                no|     |
        |                |                  |     V
        +----------------+                  +--> "all"

An example configuration that disables reservations looks as follows:

{
  "Dhcp4": {
    "subnet4": [
      {
        "id": 1,
        "pools": [
          {
            "pool": "192.0.2.10-192.0.2.100"
          }
        ],
        "reservation-mode": "disabled",
        "subnet": "192.0.2.0/24"
      }
    ]
  }
}

An example configuration using global reservations is shown below:

{
  "Dhcp4": {
    "reservation-mode": "global",
    "reservations": [
      {
        "hostname": "host-one",
        "hw-address": "01:bb:cc:dd:ee:ff"
      },
      {
        "hostname": "host-two",
        "hw-address": "02:bb:cc:dd:ee:ff"
      }
    ],
    "subnet4": [
      {
        "id": 1,
        "pools": [
          {
            "pool": "192.0.2.10-192.0.2.100"
          }
        ],
        "subnet": "192.0.2.0/24"
      }
    ]
  }
}

The meaning of the reservation flags are:

  • reservations-global: fetch global reservations.
  • reservations-in-subnet: fetch subnet reservations. For a shared network this includes all subnet members of the shared network.
  • reservations-out-of-pool: this makes sense only when the reservations-in-subnet flag is true. When reservations-out-of-pool is true, the server assumes that all host reservations are for addresses that do not belong to the dynamic pool. Therefore, it can skip the reservation checks when dealing with in-pool addresses, thus improving performance. The server will not assign reserved addresses that are inside the dynamic pools to the respective clients. This also means that the addresses matching the respective reservations from inside the dynamic pools (if any) can be dynamically assigned to any client.

The disabled value from the deprecated reservation-mode corresponds to:

{
  "Dhcp4": {
    "reservations-global": false,
    "reservations-in-subnet": false
  }
}

The global value from the deprecated reservation-mode corresponds to:

{
  "Dhcp4": {
    "reservations-global": true,
    "reservations-in-subnet": false
  }
}

The out-of-pool value from the deprecated reservation-mode corresponds to:

{
  "Dhcp4": {
    "reservations-global": false,
    "reservations-in-subnet": true,
    "reservations-out-of-pool": true
  }
}

And the all value from the deprecated reservation-mode corresponds to:

{
  "Dhcp4": {
    "reservations-global": false,
    "reservations-in-subnet": true,
    "reservations-out-of-pool": false
  }
}

To activate both global and all, the following combination can be used:

{
  "Dhcp4": {
    "reservations-global": true,
    "reservations-in-subnet": true,
    "reservations-out-of-pool": false
  }
}

To activate both global and out-of-pool, the following combination can be used:

{
  "Dhcp4": {
    "reservations-global": true,
    "reservations-in-subnet": true,
    "reservations-out-of-pool": true
  }
}

Enabling out-of-pool and disabling in-subnet at the same time is not recommended because out-of-pool applies to host reservations in a subnet, which are fetched only when the in-subnet flag is true.

The parameter can be specified at the global, subnet, and shared-network levels.

An example configuration that disables reservations looks as follows:

{
  "Dhcp4": {
    "subnet4": [
      {
        "reservations-global": false,
        "reservations-in-subnet": false,
        "subnet": "192.0.2.0/24",
        "id": 1
      }
    ]
  }
}

An example configuration using global reservations is shown below:

{
  "Dhcp4": {
    "reservations": [
      {
        "hostname": "host-one",
        "hw-address": "01:bb:cc:dd:ee:ff"
      },
      {
        "hostname": "host-two",
        "hw-address": "02:bb:cc:dd:ee:ff"
      }
    ],
    "reservations-global": true,
    "reservations-in-subnet": false,
    "subnet4": [
      {
        "pools": [
          {
            "pool": "192.0.2.10-192.0.2.100"
          }
        ],
        "subnet": "192.0.2.0/24",
        "id": 1
      }
    ]
  }
}

For more details regarding global reservations, see Global Reservations in DHCPv4.

Another aspect of host reservations is the different types of identifiers. Kea currently supports four types of identifiers: hw-address, duid, client-id, and circuit-id. This is beneficial from a usability perspective; however, there is one drawback. For each incoming packet, Kea has to extract each identifier type and then query the database to see if there is a reservation by this particular identifier. If nothing is found, the next identifier is extracted and the next query is issued. This process continues until either a reservation is found or all identifier types have been checked. Over time, with an increasing number of supported identifier types, Kea would become slower and slower.

To address this problem, a parameter called host-reservation-identifiers is available. It takes a list of identifier types as a parameter. Kea checks only those identifier types enumerated in host-reservation-identifiers. From a performance perspective, the number of identifier types should be kept to a minimum, ideally one. If the deployment uses several reservation types, please enumerate them from most- to least-frequently used, as this increases the chances of Kea finding the reservation using the fewest queries. An example of a host-reservation-identifiers configuration looks as follows:

{
"host-reservation-identifiers": [ "circuit-id", "hw-address", "duid", "client-id" ],
"subnet4": [
    {
        "subnet": "192.0.2.0/24",
        ...
    }
],
...
}

If not specified, the default value is:

"host-reservation-identifiers": [ "hw-address", "duid", "circuit-id", "client-id" ]

Note

As soon as a host reservation is found the search is stopped so when a client has two host reservations using different enabled identifier types the first is always returned and the second ignored. In other words, this is usually a a configuration mistake. In rare cases when having two reservations for the same host makes sense, you can control which of those will be used by ordering the list of identifier types in host-reservation-identifiers.

8.3.9. Global Reservations in DHCPv4

In some deployments, such as mobile, clients can roam within the network and certain parameters must be specified regardless of the client's current location. To meet such a need, Kea offers a global reservation mechanism. The idea behind it is that regular host reservations are tied to specific subnets, by using a specific subnet ID. Kea can specify a global reservation that can be used in every subnet that has global reservations enabled.

This feature can be used to assign certain parameters, such as hostname or other dedicated, host-specific options. It can also be used to assign addresses.

An address assigned via global host reservation must be feasible for the subnet the server selects for the client. In other words, the address must lie within the subnet; otherwise, it is ignored and the server will attempt to dynamically allocate an address. If the selected subnet belongs to a shared network, the server checks for feasibility against the subnet's siblings, selecting the first in-range subnet. If no such subnet exists, the server falls back to dynamically allocating the address.

Note

Prior to release 2.3.5, the server did not perform feasibility checks on globally reserved addresses, which allowed the server to be configured to hand out nonsensical leases for arbitrary address values. Later versions of Kea perform these checks.

To use global host reservations, a configuration similar to the following can be used:

"Dhcp4": {
    # This specifies global reservations.
    # They will apply to all subnets that
    # have global reservations enabled.

    "reservations": [
    {
       "hw-address": "aa:bb:cc:dd:ee:ff",
       "hostname": "hw-host-dynamic"
    },
    {
       "hw-address": "01:02:03:04:05:06",
       "hostname": "hw-host-fixed",

       # Use of IP addresses in global reservations is risky.
       # If used outside of a matching subnet, such as 192.0.1.0/24,
       # it will result in a broken configuration being handed
       # to the client.
       "ip-address": "192.0.1.77"
    },
    {
       "duid": "01:02:03:04:05",
       "hostname": "duid-host"
    },
    {
       "circuit-id": "'charter950'",
       "hostname": "circuit-id-host"
    },
    {
       "client-id": "01:11:22:33:44:55:66",
       "hostname": "client-id-host"
    }
    ],
    "valid-lifetime": 600,
    "subnet4": [
    {
        "id": 1,
        "subnet": "10.0.0.0/24",
        # It is replaced by the "reservations-global",
        # "reservations-in-subnet", and "reservations-out-of-pool"
        # parameters.
        # "reservation-mode": "global",
        # Specify if the server should look up global reservations.
        "reservations-global": true,
        # Specify if the server should look up in-subnet reservations.
        "reservations-in-subnet": false,
        # Specify if the server can assume that all reserved addresses
        # are out-of-pool. It can be ignored because "reservations-in-subnet"
        # is false.
        # "reservations-out-of-pool": false,
        "pools": [ { "pool": "10.0.0.10-10.0.0.100" } ]
    }
    ]
}

When using database backends, the global host reservations are distinguished from regular reservations by using a subnet-id value of 0.

8.3.10. Pool Selection with Client Class Reservations

Client classes can be specified in the Kea configuration file and/or via host reservations. The classes specified in the Kea configuration file are evaluated immediately after receiving the DHCP packet and therefore can be used to influence subnet selection using the client-class parameter specified in the subnet scope. The classes specified within the host reservations are fetched and assigned to the packet after the server has already selected a subnet for the client. This means that the client class specified within a host reservation cannot be used to influence subnet assignment for this client, unless the subnet belongs to a shared network. If the subnet belongs to a shared network, the server may dynamically change the subnet assignment while trying to allocate a lease. If the subnet does not belong to a shared network, the subnet is not changed once selected.

If the subnet does not belong to a shared network, it is possible to use host reservation-based client classification to select an address pool within the subnet as follows:

"Dhcp4": {
    "client-classes": [
        {
            "name": "reserved_class"
        },
        {
            "name": "unreserved_class",
            "test": "not member('reserved_class')"
        }
    ],
    "subnet4": [
        {
            "id": 1,
            "subnet": "192.0.2.0/24",
            "reservations": [
                {
                    "hw-address": "aa:bb:cc:dd:ee:fe",
                    "client-classes": [ "reserved_class" ]
                }
            ],
            "pools": [
                {
                    "pool": "192.0.2.10-192.0.2.20",
                    "client-class": "reserved_class"
                },
                {
                    "pool": "192.0.2.30-192.0.2.40",
                    "client-class": "unreserved_class"
                }
            ]
        }
    ]
}

The reserved_class is declared without the test parameter because it may only be assigned to the client via the host reservation mechanism. The second class, unreserved_class, is assigned to clients which do not belong to the reserved_class. The first pool within the subnet is only used for clients having a reservation for the reserved_class. The second pool is used for clients not having such a reservation. The configuration snippet includes one host reservation which causes the client with the MAC address aa:bb:cc:dd:ee:fe to be assigned to the reserved_class. Thus, this client will be given an IP address from the first address pool.

8.3.11. Subnet Selection with Client Class Reservations

There is one specific use case when subnet selection may be influenced by client classes specified within host reservations: when the client belongs to a shared network. In such a case it is possible to use classification to select a subnet within this shared network. Consider the following example:

"Dhcp4": {
    "client-classes": [
        {
            "name": "reserved_class"
        },
        {
            "name": "unreserved_class",
            "test": "not member('reserved_class')"
        }
    ],
    "reservations": [
        {
            "hw-address": "aa:bb:cc:dd:ee:fe",
            "client-classes": [ "reserved_class" ]
        }
    ],
    # It is replaced by the "reservations-global",
    # "reservations-in-subnet", and "reservations-out-of-pool" parameters.
    # Specify if the server should look up global reservations.
    "reservations-global": true,
    # Specify if the server should look up in-subnet reservations.
    "reservations-in-subnet": false,
    # Specify if the server can assume that all reserved addresses
    # are out-of-pool. It can be ignored because "reservations-in-subnet"
    # is false, but if specified, it is inherited by "shared-networks"
    # and "subnet4" levels.
    # "reservations-out-of-pool": false,
    "shared-networks": [
        {
        "subnet4": [
            {
                "id": 1,
                "subnet": "192.0.2.0/24",
                "pools": [
                    {
                        "pool": "192.0.2.10-192.0.2.20",
                        "client-class": "reserved_class"
                    }
                ]
            },
            {
                "id": 2,
                "subnet": "192.0.3.0/24",
                "pools": [
                    {
                        "pool": "192.0.3.10-192.0.3.20",
                        "client-class": "unreserved_class"
                    }
                ]
            }
        ]
        }
    ]
}

This is similar to the example described in Pool Selection with Client Class Reservations. This time, however, there are two subnets, each of which has a pool associated with a different class. The clients that do not have a reservation for the reserved_class are assigned an address from the subnet 192.0.3.0/24. Clients with a reservation for the reserved_class are assigned an address from the subnet 192.0.2.0/24. The subnets must belong to the same shared network. In addition, the reservation for the client class must be specified at the global scope (global reservation) and reservations-global must be set to true.

In the example above, the client-class could also be specified at the subnet level rather than the pool level, and would yield the same effect.

8.3.12. Multiple Reservations for the Same IP

Host reservations were designed to preclude the creation of multiple reservations for the same IP address within a particular subnet, to avoid having two different clients compete for the same address. When using the default settings, the server returns a configuration error when it finds two or more reservations for the same IP address within a subnet in the Kea configuration file. libdhcp_host_cmds.so returns an error in response to the reservation-add command when it detects that the reservation exists in the database for the IP address for which the new reservation is being added.

In some deployments a single host can select one of several network interfaces to communicate with the DHCP server, and the server must assign the same IP address to the host regardless of the interface used. Since each interface is assigned a different MAC address, it implies that several host reservations must be created to associate all of the MAC addresses present on this host with IP addresses. Using different IP addresses for each interface is impractical and is considered a waste of the IPv4 address space, especially since the host typically uses only one interface for communication with the server, hence only one IP address is in use.

This causes a need to create multiple host reservations for a single IP address within a subnet; this is supported since the Kea 1.9.1 release as an optional mode of operation, enabled with the ip-reservations-unique global parameter.

The ip-reservations-unique is a boolean parameter that defaults to true, which forbids the specification of more than one reservation for the same IP address within a given subnet. Setting this parameter to false allows such reservations to be created both in the Kea configuration file and in the host database backend, via libdhcp_host_cmds.so.

Setting ip-reservations-unique to false when using memfile, MySQL or PostgreSQL is supported. This setting is not supported when using Host Cache (see libdhcp_host_cache.so: Host Cache Reservations for Improved Performance), and the RADIUS backend (see libdhcp_radius.so: RADIUS Server Support). These reservation backends simply do not support multiple reservations for the same IP. If either of these hooks are loaded and ip-reservations-unique is set to false, then a configuration error will be emitted and the server will fail to start.

Note

When ip-reservations-unique is set to true (the default value), the server ensures that IP reservations are unique for a subnet within a single host backend and/or Kea configuration file. It does not guarantee that the reservations are unique across multiple backends.

The following is an example configuration with two reservations for the same IP address but different MAC addresses:

"Dhcp4": {
    "ip-reservations-unique": false,
    "subnet4": [
        {
            "id": 1,
            "subnet": "192.0.2.0/24",
            "reservations": [
                {
                    "hw-address": "1a:1b:1c:1d:1e:1f",
                    "ip-address": "192.0.2.11"
                },
                {
                    "hw-address": "2a:2b:2c:2d:2e:2f",
                    "ip-address": "192.0.2.11"
                }
            ]
        }
    ]
}

It is possible to control the ip-reservations-unique parameter via the Configuration Backend in DHCPv4. If the new setting of this parameter conflicts with the currently used backends (i.e. backends do not support the new setting), the new setting is ignored and a warning log message is generated. The backends continue to use the default setting, expecting that IP reservations are unique within each subnet. To allow the creation of non-unique IP reservations, the administrator must remove the backends which lack support for them from the configuration file.

Administrators must be careful when they have been using multiple reservations for the same IP address and later decide to return to the default mode in which this is no longer allowed. They must make sure that at most one reservation for a given IP address exists within a subnet, prior to switching back to the default mode. If such duplicates are left in the configuration file, the server reports a configuration error. Leaving such reservations in the host databases does not cause configuration errors but may lead to lease allocation errors during the server's operation, when it unexpectedly finds multiple reservations for the same IP address.

Note

Currently the Kea server does not verify whether multiple reservations for the same IP address exist in MySQL and/or PostgreSQL host databases when ip-reservations-unique is updated from true to false. This may cause issues with lease allocations. The administrator must ensure that there is at most one reservation for each IP address within each subnet, prior to the configuration update.

The reservations-lookup-first is a boolean parameter which controls whether host reservations lookup should be performed before lease lookup. This parameter has effect only when multi-threading is disabled. When multi-threading is enabled, host reservations lookup is always performed first to avoid lease-lookup resource locking. The reservations-lookup-first parameter defaults to false when multi-threading is disabled.

8.3.13. Host Reservations as Basic Access Control

Starting with Kea 2.3.5, it is possible to define a host reservation that contains just an identifier, without any address, options, or values. In some deployments this is useful, as the hosts that have a reservation belong to the KNOWN class while others do not. This can be used as a basic access control mechanism.

The following example demonstrates this concept. It indicates a single IPv4 subnet and all clients will get an address from it. However, only known clients (those that have reservations) will get their default router configured. Empty reservations i.e. reservations that only have the identification criterion, can be specifically useful in this regard of making the clients known.

"Dhcp4": {
    "client-classes": [
        {
            "name": "KNOWN",
            "option-data": [
                {
                    "name": "routers",
                    "data": "192.0.2.250"
                }
            ]
        }
    ],
    "reservations": [
        // Clients on this list will be added to the KNOWN class.
        { "hw-address": "aa:bb:cc:dd:ee:fe" },
        { "hw-address": "11:22:33:44:55:66" }
    ],
    "reservations-in-subnet": true,

    "subnet4": [
        {
            "id": 1,
            "subnet": "192.0.2.0/24",
            "pools": [
                {
                    "pool": "192.0.2.1-192.0.2.200"
                }
            ]
        }
    ]
}

This concept can be extended further. A good real-life scenario might be a situation where some customers of an ISP have not paid their bills. A new class can be defined to use an alternative default router that, instead of relaying traffic, redirects those customers to a captive portal urging them to bring their accounts up to date.

"Dhcp4": {
    "client-classes": [
        {
            "name": "blocked",
            "option-data": [
                {
                    "name": "routers",
                    "data": "192.0.2.251"
                }
            ]
        }
    ],
    "reservations": [
        // Clients on this list will be added to the KNOWN class. Some
        // will also be added to the blocked class.
        { "hw-address": "aa:bb:cc:dd:ee:fe",
          "client-classes": [ "blocked" ] },
        { "hw-address": "11:22:33:44:55:66" }
    ],
    "reservations-in-subnet": true,

    "subnet4": [
        {
            "id": 1,
            "subnet": "192.0.2.0/24",
            "pools": [
                {
                    "pool": "192.0.2.1-192.0.2.200"
                }
            ],
            "option-data": [
                {
                    "name": "routers",
                    "data": "192.0.2.250"
                }
            ]
        }
    ]
}

8.4. Shared Networks in DHCPv4

DHCP servers use subnet information in two ways. It is used to both determine the point of attachment, i.e. where the client is connected to the network, and to group information pertaining to a specific location in the network. Sometimes it is useful to have more than one logical IP subnet deployed on the same physical link. Understanding that two or more subnets are used on the same link requires additional logic in the DHCP server. This capability is called "shared networks" in Kea, and sometimes also "shared subnets"; in Microsoft's nomenclature it is called "multinet."

There are many cases where the shared networks feature is useful; here we explain just a handful of the most common ones. The first and by far most common use case is an existing IPv4 network that has grown and is running out of available address space. Rather than migrating all devices to a new, larger subnet, it is easier to simply configure additional subnets on top of the existing one. Sometimes, due to address space fragmentation (e.g. only many disjointed /24s are available), this is the only choice. Also, configuring additional subnets has the advantage of not disrupting the operation of existing devices.

Another very frequent use case comes from cable networks. There are two types of devices in cable networks: cable modems and the end-user devices behind them. It is a common practice to use different subnets for cable modems to prevent users from tinkering with them. In this case, the distinction is based on the type of device, rather than on address-space exhaustion.

A client connected to a shared network may be assigned an address from any of the pools defined within the subnets belonging to the shared network. Internally, the server selects one of the subnets belonging to a shared network and tries to allocate an address from this subnet. If the server is unable to allocate an address from the selected subnet (e.g., due to address-pool exhaustion), it uses another subnet from the same shared network and tries to allocate an address from this subnet. The server typically allocates all addresses available in a given subnet before it starts allocating addresses from other subnets belonging to the same shared network. However, in certain situations the client can be allocated an address from another subnet before the address pools in the first subnet get exhausted; this sometimes occurs when the client provides a hint that belongs to another subnet, or the client has reservations in a subnet other than the default.

Note

Deployments should not assume that Kea waits until it has allocated all the addresses from the first subnet in a shared network before allocating addresses from other subnets.

To define a shared network, an additional configuration scope is introduced:

{
"Dhcp4": {
    "shared-networks": [
        {
            # Name of the shared network. It may be an arbitrary string
            # and it must be unique among all shared networks.
            "name": "my-secret-lair-level-1",

            # The subnet selector can be specified at the shared-network level.
            # Subnets from this shared network will be selected for directly
            # connected clients sending requests to the server's "eth0" interface.
            "interface": "eth0",

            # This starts a list of subnets in this shared network.
            # There are two subnets in this example.
            "subnet4": [
                {
                    "id": 1,
                    "subnet": "10.0.0.0/8",
                    "pools": [ { "pool":  "10.0.0.1 - 10.0.0.99" } ]
                },
                {
                    "id": 2,
                    "subnet": "192.0.2.0/24",
                    "pools": [ { "pool":  "192.0.2.100 - 192.0.2.199" } ]
                }
            ]
        }
    ],
    # end of shared-networks

    # It is likely that in the network there will be a mix of regular,
    # "plain" subnets and shared networks. It is perfectly valid to mix
    # them in the same configuration file.
    #
    # This is a regular subnet. It is not part of any shared network.
    "subnet4": [
        {
            "id": 3,
            "subnet": "192.0.3.0/24",
            "pools": [ { "pool":  "192.0.3.1 - 192.0.3.200" } ],
            "interface": "eth1"
        }
    ]
}
}

As demonstrated in the example, it is possible to mix shared and regular ("plain") subnets. Each shared network must have a unique name. This is similar to the ID for subnets, but gives administrators more flexibility. It is used for logging, but also internally for identifying shared networks.

In principle it makes sense to define only shared networks that consist of two or more subnets. However, for testing purposes, an empty subnet or a network with just a single subnet is allowed. This is not a recommended practice in production networks, as the shared network logic requires additional processing and thus lowers the server's performance. To avoid unnecessary performance degradation, shared subnets should only be defined when required by the deployment.

Shared networks provide the ability to specify many parameters in the shared network scope that apply to all subnets within it. If necessary, it is possible to specify a parameter in the shared-network scope and then override its value in the subnet scope. For example:

{
"shared-networks": [
    {
        "name": "lab-network3",

        "interface": "eth0",

        # This applies to all subnets in this shared network, unless
        # values are overridden on subnet scope.
        "valid-lifetime": 600,

        # This option is made available to all subnets in this shared
        # network.
        "option-data": [
        {
            "name": "log-servers",
            "data": "1.2.3.4"
        }
        ],

        "subnet4": [
            {
                "id": 1,
                "subnet": "10.0.0.0/8",
                "pools": [ { "pool":  "10.0.0.1 - 10.0.0.99" } ],

                # This particular subnet uses different values.
                "valid-lifetime": 1200,
                "option-data": [
                {
                    "name": "log-servers",
                    "data": "10.0.0.254"
                },
                {
                    "name": "routers",
                    "data": "10.0.0.254"
                } ]
            },
            {
                 "id": 2,
                 "subnet": "192.0.2.0/24",
                 "pools": [ { "pool":  "192.0.2.100 - 192.0.2.199" } ],

                 # This subnet does not specify its own valid-lifetime value,
                 # so it is inherited from shared network scope.
                 "option-data": [
                 {
                     "name": "routers",
                     "data": "192.0.2.1"
                 } ]
            }
        ]
    }
    ],
    ...
}

In this example, there is a log-servers option defined that is available to clients in both subnets in this shared network. Also, the valid lifetime is set to 10 minutes (600s). However, the first subnet overrides some of the values (the valid lifetime is 20 minutes, there is a different IP address for log-servers), but also adds its own option (the router address). Assuming a client asking for router and log-servers options is assigned a lease from this subnet, it will get a lease for 20 minutes and a log-servers and routers value of 10.0.0.254. If the same client is assigned to the second subnet, it will get a 10-minute lease, a log-servers value of 1.2.3.4, and routers set to 192.0.2.1.

8.4.1. Local and Relayed Traffic in Shared Networks

It is possible to specify an interface name at the shared-network level, to tell the server that this specific shared network is reachable directly (not via relays) using the local network interface. As all subnets in a shared network are expected to be used on the same physical link, it is a configuration error to attempt to define a shared network using subnets that are reachable over different interfaces. In other words, all subnets within the shared network must have the same value for the interface parameter. The following configuration is an example of what NOT to do:

{
"shared-networks": [
    {
        "name": "office-floor-2",
        "subnet4": [
            {
                "id": 1,
                "subnet": "10.0.0.0/8",
                "pools": [ { "pool":  "10.0.0.1 - 10.0.0.99" } ],
                "interface": "eth0"
            },
            {
                 "id": 2,
                 "subnet": "192.0.2.0/24",
                 "pools": [ { "pool":  "192.0.2.100 - 192.0.2.199" } ],

                 # Specifying the different interface name is a configuration
                 # error. This value should rather be "eth0" or the interface
                 # name in the other subnet should be "eth1".
                 "interface": "eth1"
            }
        ]
    }
],
...
}

To minimize the chance of configuration errors, it is often more convenient to simply specify the interface name once, at the shared-network level, as shown in the example below.

{
"shared-networks": [
    {
        "name": "office-floor-2",

        # This tells Kea that the whole shared network is reachable over a
        # local interface. This applies to all subnets in this network.
        "interface": "eth0",

        "subnet4": [
            {
                "id": 1,
                "subnet": "10.0.0.0/8",
                "pools": [ { "pool":  "10.0.0.1 - 10.0.0.99" } ]
            },
            {
                 "id": 2,
                 "subnet": "192.0.2.0/24",
                 "pools": [ { "pool":  "192.0.2.100 - 192.0.2.199" } ]
            }
        ]
    }
],
...
}

With relayed traffic, subnets are typically selected using the relay agents' addresses. If the subnets are used independently (not grouped within a shared network), a different relay address can be specified for each of these subnets. When multiple subnets belong to a shared network they must be selected via the same relay address and, similarly to the case of the local traffic described above, it is a configuration error to specify different relay addresses for the respective subnets in the shared network. The following configuration is another example of what NOT to do:

{
"shared-networks": [
    {
        "name": "kakapo",
        "subnet4": [
            {
                "id": 1,
                "subnet": "192.0.2.0/26",
                "relay": {
                    "ip-addresses": [ "192.1.1.1" ]
                },
                "pools": [ { "pool": "192.0.2.63 - 192.0.2.63" } ]
            },
            {
                "id": 2,
                "subnet": "10.0.0.0/24",
                "relay": {
                    # Specifying a different relay address for this
                    # subnet is a configuration error. In this case
                    # it should be 192.1.1.1 or the relay address
                    # in the previous subnet should be 192.2.2.2.
                    "ip-addresses": [ "192.2.2.2" ]
                },
                "pools": [ { "pool": "10.0.0.16 - 10.0.0.16" } ]
            }
        ]
    }
],
...
}

Again, it is better to specify the relay address at the shared-network level; this value will be inherited by all subnets belonging to the shared network.

{
"shared-networks": [
    {
        "name": "kakapo",
        "relay": {
            # This relay address is inherited by both subnets.
            "ip-addresses": [ "192.1.1.1" ]
        },
        "subnet4": [
            {
                "id": 1,
                "subnet": "192.0.2.0/26",
                "pools": [ { "pool": "192.0.2.63 - 192.0.2.63" } ]
            },
            {
                "id": 2,
                "subnet": "10.0.0.0/24",
                "pools": [ { "pool": "10.0.0.16 - 10.0.0.16" } ]
            }
        ]
    }
],
...
}

Even though it is technically possible to configure two (or more) subnets within the shared network to use different relay addresses, this will almost always lead to a different behavior than what the user expects. In this case, the Kea server will initially select one of the subnets by matching the relay address in the client's packet with the subnet's configuration. However, it MAY end up using the other subnet (even though it does not match the relay address) if the client already has a lease in this subnet or has a host reservation in this subnet, or simply if the initially selected subnet has no more addresses available. Therefore, it is strongly recommended to always specify subnet selectors (interface or relay address) at the shared-network level if the subnets belong to a shared network, as it is rarely useful to specify them at the subnet level and may lead to the configuration errors described above.

8.4.2. Client Classification in Shared Networks

Sometimes it is desirable to segregate clients into specific subnets based on certain properties. This mechanism is called client classification and is described in Client Classification. Client classification can be applied to subnets belonging to shared networks in the same way as it is used for subnets specified outside of shared networks. It is important to understand how the server selects subnets for clients when client classification is in use, to ensure that the appropriate subnet is selected for a given client type.

If a subnet is associated with a class, only the clients belonging to this class can use this subnet. If there are no classes specified for a subnet, any client connected to a given shared network can use this subnet. A common mistake is to assume that a subnet that includes a client class is preferred over subnets without client classes. Consider the following example:

{
    "client-classes": [
        {
            "name": "b-devices",
            "test": "option[93].hex == 0x0002"
        }
    ],
    "shared-networks": [
        {
            "name": "galah",
            "interface": "eth0",
            "subnet4": [
                {
                    "id": 1,
                    "subnet": "192.0.2.0/26",
                    "pools": [ { "pool": "192.0.2.1 - 192.0.2.63" } ]
                },
                {
                    "id": 2,
                    "subnet": "10.0.0.0/24",
                    "pools": [ { "pool": "10.0.0.2 - 10.0.0.250" } ],
                    "client-class": "b-devices"
                }
            ]
        }
    ]
}

If the client belongs to the "b-devices" class (because it includes option 93 with a value of 0x0002), that does not guarantee that the subnet 10.0.0.0/24 will be used (or preferred) for this client. The server can use either of the two subnets, because the subnet 192.0.2.0/26 is also allowed for this client. The client classification used in this case should be perceived as a way to restrict access to certain subnets, rather than as a way to express subnet preference. For example, if the client does not belong to the "b-devices" class, it may only use the subnet 192.0.2.0/26 and will never use the subnet 10.0.0.0/24.

A typical use case for client classification is in a cable network, where cable modems should use one subnet and other devices should use another subnet within the same shared network. In this case it is necessary to apply classification on all subnets. The following example defines two classes of devices, and the subnet selection is made based on option 93 values.

{
    "client-classes": [
        {

            "name": "a-devices",
            "test": "option[93].hex == 0x0001"
        },
        {
            "name": "b-devices",
            "test": "option[93].hex == 0x0002"
        }
    ],
    "shared-networks": [
        {
            "name": "galah",
            "interface": "eth0",
            "subnet4": [
                {
                    "id": 1,
                    "subnet": "192.0.2.0/26",
                    "pools": [ { "pool": "192.0.2.1 - 192.0.2.63" } ],
                    "client-class": "a-devices"
                },
                {
                    "id": 2,
                    "subnet": "10.0.0.0/24",
                    "pools": [ { "pool": "10.0.0.2 - 10.0.0.250" } ],
                    "client-class": "b-devices"
                }
            ]
        }
    ]
}

In this example each class has its own restriction. Only clients that belong to class "a-devices" are able to use subnet 192.0.2.0/26 and only clients belonging to "b-devices" are able to use subnet 10.0.0.0/24. Care should be taken not to define too-restrictive classification rules, as clients that are unable to use any subnets will be refused service. However, this may be a desired outcome if one wishes to provide service only to clients with known properties (e.g. only VoIP phones allowed on a given link).

It is possible to achieve an effect similar to the one presented in this section without the use of shared networks. If the subnets are placed in the global subnets scope, rather than in the shared network, the server will still use classification rules to pick the right subnet for a given class of devices. The major benefit of placing subnets within the shared network is that common parameters for the logically grouped subnets can be specified once in the shared-network scope, e.g. the interface or relay parameter. All subnets belonging to this shared network will inherit those parameters.

8.4.3. Host Reservations in Shared Networks

Subnets that are part of a shared network allow host reservations, similar to regular subnets:

{
    "shared-networks": [
    {
        "name": "frog",
        "interface": "eth0",
        "subnet4": [
            {
                "subnet": "192.0.2.0/26",
                "id": 100,
                "pools": [ { "pool": "192.0.2.1 - 192.0.2.63" } ],
                "reservations": [
                    {
                        "hw-address": "aa:bb:cc:dd:ee:ff",
                        "ip-address": "192.0.2.28"
                    }
                ]
            },
            {
                "subnet": "10.0.0.0/24",
                "id": 101,
                "pools": [ { "pool": "10.0.0.1 - 10.0.0.254" } ],
                "reservations": [
                    {
                        "hw-address": "11:22:33:44:55:66",
                        "ip-address": "10.0.0.29"
                    }
                ]
            }
        ]
    }
    ]
}

It is worth noting that Kea conducts additional checks when processing a packet if shared networks are defined. First, instead of simply checking whether there is a reservation for a given client in its initially selected subnet, Kea looks through all subnets in a shared network for a reservation. This is one of the reasons why defining a shared network may impact performance. If there is a reservation for a client in any subnet, that particular subnet is selected for the client. Although it is technically not an error, it is considered bad practice to define reservations for the same host in multiple subnets belonging to the same shared network.

While not strictly mandatory, it is strongly recommended to use explicit "id" values for subnets if database storage will be used for host reservations. If an ID is not specified, the values for it are auto-generated, i.e. Kea assigns increasing integer values starting from 1. Thus, the auto-generated IDs are not stable across configuration changes.

8.5. Server Identifier in DHCPv4

The DHCPv4 protocol uses a "server identifier" to allow clients to discriminate between several servers present on the same link; this value is an IPv4 address of the server. The server chooses the IPv4 address of the interface on which the message from the client (or relay) has been received. A single server instance uses multiple server identifiers if it is receiving queries on multiple interfaces.

It is possible to override the default server identifier values by specifying the dhcp-server-identifier option. This option configuration is only supported at the subnet, shared network, client class, and global levels. It must not be specified at the host-reservation level. When configuring the dhcp-server-identifier option at client-class level, the class must not set the only-if-required flag, because this class would not be evaluated before the server determines if the received DHCP message should be accepted for processing. Such classes are evaluated after subnet selection. See Required Classification for details.

The following example demonstrates how to override the server identifier for a subnet:

{
"subnet4": [
    {
        "subnet": "192.0.2.0/24",
        "option-data": [
            {
                "name": "dhcp-server-identifier",
                "data": "10.2.5.76"
            }
        ],
        ...
    }
],
...
}

8.6. How the DHCPv4 Server Selects a Subnet for the Client

The DHCPv4 server differentiates among directly connected clients, clients trying to renew leases, and clients sending their messages through relays. For directly connected clients, the server checks the configuration for the interface on which the message has been received and, if the server configuration does not match any configured subnet, the message is discarded.

An optional interface parameter is available within a subnet definition to designate that a given subnet is local, i.e. reachable directly over the specified interface. For example, a server that is intended to serve a local subnet over eth0 may be configured as follows:

"Dhcp4": {
    "subnet4": [
        {
            "id": 1,
            "subnet": "192.0.2.0/24",
            "pools": [
                 {
                     "pool": "192.0.2.100 - 192.0.2.199"
                 }
             ],
            "interface": "eth0"
        }
    ],
    ...
}

Assuming that the server's interface is configured with the IPv4 address 192.0.2.3, the server only processes messages received through this interface from a directly connected client if there is a subnet configured to which this IPv4 address belongs, such as 192.0.2.0/24. The server uses this subnet to assign an IPv4 address for the client.

The rule above does not apply when the client unicasts its message, i.e. is trying to renew its lease; such a message is accepted through any interface. The renewing client sets ciaddr to the currently used IPv4 address, and the server uses this address to select the subnet for the client (in particular, to extend the lease using this address).

If the message is relayed it is accepted through any interface. The giaddr set by the relay agent is used to select the subnet for the client.

It is also possible to specify a relay IPv4 address for a given subnet. It can be used to match incoming packets into a subnet in uncommon configurations, e.g. shared networks. See Using a Specific Relay Agent for a Subnet for details.

Note

The subnet selection mechanism described in this section is based on the assumption that client classification is not used. The classification mechanism alters the way in which a subnet is selected for the client, depending on the classes to which the client belongs.

Note

When the selected subnet is a member of a shared network, the whole shared network is selected.

8.6.1. Using a Specific Relay Agent for a Subnet

A relay must have an interface connected to the link on which the clients are being configured. Typically the relay has an IPv4 address configured on that interface, which belongs to the subnet from which the server assigns addresses. Normally, the server is able to use the IPv4 address inserted by the relay (in the giaddr field of the DHCPv4 packet) to select the appropriate subnet.

However, that is not always the case. In certain uncommon — but valid — deployments, the relay address may not match the subnet. This usually means that there is more than one subnet allocated for a given link. The two most common examples of this are long-lasting network renumbering (where both old and new address spaces are still being used) and a cable network. In a cable network, both cable modems and the devices behind them are physically connected to the same link, yet they use distinct addressing. In such a case, the DHCPv4 server needs additional information (the IPv4 address of the relay) to properly select an appropriate subnet.

The following example assumes that there is a subnet 192.0.2.0/24 that is accessible via a relay that uses 10.0.0.1 as its IPv4 address. The server is able to select this subnet for any incoming packets that come from a relay that has an address in the 192.0.2.0/24 subnet. It also selects that subnet for a relay with address 10.0.0.1.

{
  "Dhcp4": {
    "subnet4": [
        {
            "id": 1,
            "subnet": "192.0.2.0/24",
            "pools": [ { "pool": "192.0.2.10 - 192.0.2.20" } ],
            "relay": {
                "ip-addresses": [ "10.0.0.1" ]
            }
        }
    ]
  }
}

If relay is specified, the ip-addresses parameter within it is mandatory. The ip-addresses parameter supports specifying a list of addresses.

8.6.2. Segregating IPv4 Clients in a Cable Network

In certain cases, it is useful to mix relay address information (introduced in Using a Specific Relay Agent for a Subnet) with client classification (explained in Client Classification). One specific example is in a cable network, where modems typically get addresses from a different subnet than all the devices connected behind them.

Let us assume that there is one Cable Modem Termination System (CMTS) with one CM MAC (a physical link that modems are connected to). We want the modems to get addresses from the 10.1.1.0/24 subnet, while everything connected behind the modems should get addresses from the 192.0.2.0/24 subnet. The CMTS that acts as a relay uses address 10.1.1.1. The following configuration can serve that situation:

"Dhcp4": {
    "subnet4": [
        {
            "id": 1,
            "subnet": "10.1.1.0/24",
            "pools":  [ { "pool": "10.1.1.2 - 10.1.1.20" } ],
            "client-class": "docsis3.0",
            "relay": {
                "ip-addresses": [ "10.1.1.1" ]
            }
        },
        {
            "id": 2,
            "subnet": "192.0.2.0/24",
            "pools": [ { "pool": "192.0.2.10 - 192.0.2.20" } ],
            "relay": {
                "ip-addresses": [ "10.1.1.1" ]
            }
        }
    ],
    ...
}

8.7. Duplicate Addresses (DHCPDECLINE Support)

The DHCPv4 server is configured with a certain pool of addresses that it is expected to hand out to DHCPv4 clients. It is assumed that the server is authoritative and has complete jurisdiction over those addresses. However, for various reasons such as misconfiguration or a faulty client implementation that retains its address beyond the valid lifetime, there may be devices connected that use those addresses without the server's approval or knowledge.

Such an unwelcome event can be detected by legitimate clients (using ARP or ICMP Echo Request mechanisms) and reported to the DHCPv4 server using a DHCPDECLINE message. The server does a sanity check (to see whether the client declining an address really was supposed to use it) and then conducts a clean-up operation. Any DNS entries related to that address are removed, the event is logged, and hooks are triggered. After that is complete, the address is marked as declined (which indicates that it is used by an unknown entity and thus not available for assignment) and a probation time is set on it. Unless otherwise configured, the probation period lasts 24 hours; after that time, the server will recover the lease (i.e. put it back into the available state) and the address will be available for assignment again. It should be noted that if the underlying issue of a misconfigured device is not resolved, the duplicate-address scenario will repeat. If reconfigured correctly, this mechanism provides an opportunity to recover from such an event automatically, without any system administrator intervention.

To configure the decline probation period to a value other than the default, the following syntax can be used:

"Dhcp4": {
    "decline-probation-period": 3600,
    "subnet4": [
        {
            ...
        },
        ...
    ],
    ...
}

The parameter is expressed in seconds, so the example above instructs the server to recycle declined leases after one hour.

There are several statistics and hook points associated with the decline handling procedure. The lease4_decline hook point is triggered after the incoming DHCPDECLINE message has been sanitized and the server is about to decline the lease. The declined-addresses statistic is increased after the hook returns (both the global and subnet-specific variants). (See Statistics in the DHCPv4 Server and Hook Libraries for more details on DHCPv4 statistics and Kea hook points.)

Once the probation time elapses, the declined lease is recovered using the standard expired-lease reclamation procedure, with several additional steps. In particular, both declined-addresses statistics (global and subnet-specific) are decreased. At the same time, reclaimed-declined-addresses statistics (again in two variants, global and subnet-specific) are increased.

A note about statistics: The Kea server does not decrease the assigned-addresses statistics when a DHCPDECLINE is received and processed successfully. While technically a declined address is no longer assigned, the primary usage of the assigned-addresses statistic is to monitor pool utilization. Most people would forget to include declined-addresses in the calculation, and would simply use assigned-addresses/total-addresses. This would cause a bias towards under-representing pool utilization. As this has a potential to cause serious confusion, ISC decided not to decrease assigned-addresses immediately after receiving DHCPDECLINE, but to do it later when Kea recovers the address back to the available pool.

8.8. Statistics in the DHCPv4 Server

The DHCPv4 server supports the following statistics:

DHCPv4 statistics
Statistic Data Type Description
pkt4-received integer Number of DHCPv4 packets received. This includes all packets: valid, bogus, corrupted, rejected, etc. This statistic is expected to grow rapidly.
pkt4-discover-received integer Number of DHCPDISCOVER packets received. This statistic is expected to grow; its increase means that clients that just booted started their configuration process and their initial packets reached the Kea server.
pkt4-offer-received integer Number of DHCPOFFER packets received. This statistic is expected to remain zero at all times, as DHCPOFFER packets are sent by the server and the server is never expected to receive them. A non-zero value indicates an error. One likely cause would be a misbehaving relay agent that incorrectly forwards DHCPOFFER messages towards the server, rather than back to the clients.
pkt4-request-received integer Number of DHCPREQUEST packets received. This statistic is expected to grow. Its increase means that clients that just booted received the server's response (DHCPOFFER) and accepted it, and are now requesting an address (DHCPREQUEST).
pkt4-ack-received integer Number of DHCPACK packets received. This statistic is expected to remain zero at all times, as DHCPACK packets are sent by the server and the server is never expected to receive them. A non-zero value indicates an error. One likely cause would be a misbehaving relay agent that incorrectly forwards DHCPACK messages towards the server, rather than back to the clients.
pkt4-nak-received integer Number of DHCPNAK packets received. This statistic is expected to remain zero at all times, as DHCPNAK packets are sent by the server and the server is never expected to receive them. A non-zero value indicates an error. One likely cause would be a misbehaving relay agent that incorrectly forwards DHCPNAK messages towards the server, rather than back to the clients.
pkt4-release-received integer Number of DHCPRELEASE packets received. This statistic is expected to grow. Its increase means that clients that had an address are shutting down or ceasing to use their addresses.
pkt4-decline-received integer Number of DHCPDECLINE packets received. This statistic is expected to remain close to zero. Its increase means that a client leased an address, but discovered that the address is currently used by an unknown device elsewhere in the network.
pkt4-inform-received integer Number of DHCPINFORM packets received. This statistic is expected to grow. Its increase means that there are clients that either do not need an address or already have an address and are interested only in getting additional configuration parameters.
pkt4-unknown-received integer Number of packets received of an unknown type. A non-zero value of this statistic indicates that the server received a packet that it was not able to recognize, either with an unsupported type or possibly malformed (without a message-type option).
pkt4-sent integer Number of DHCPv4 packets sent. This statistic is expected to grow every time the server transmits a packet. In general, it should roughly match pkt4-received, as most incoming packets cause the server to respond. There are exceptions (e.g. DHCPRELEASE), so do not worry if it is less than pkt4-received.
pkt4-offer-sent integer Number of DHCPOFFER packets sent. This statistic is expected to grow in most cases after a DHCPDISCOVER is processed. There are certain uncommon, but valid, cases where incoming DHCPDISCOVER packets are dropped, but in general this statistic is expected to be close to pkt4-discover-received.
pkt4-ack-sent integer Number of DHCPACK packets sent. This statistic is expected to grow in most cases after a DHCPREQUEST is processed. There are certain cases where DHCPNAK is sent instead. In general, the sum of pkt4-ack-sent and pkt4-nak-sent should be close to pkt4-request-received.
pkt4-nak-sent integer Number of DHCPNAK packets sent. This statistic is expected to grow when the server chooses not to honor the address requested by a client. In general, the sum of pkt4-ack-sent and pkt4-nak-sent should be close to pkt4-request-received.
pkt4-parse-failed integer Number of incoming packets that could not be parsed. A non-zero value of this statistic indicates that the server received a malformed or truncated packet. This may indicate problems in the network, faulty clients, or a bug in the server.
pkt4-receive-drop integer Number of incoming packets that were dropped. The exact reason for dropping packets is logged, but the most common reasons may be: an unacceptable packet type was received, direct responses are forbidden, or the server-id sent by the client does not match the server's server-id.
subnet[id].total-addresses integer Total number of addresses available for DHCPv4 management for a given subnet; in other words, this is the count of all addresses in all configured pools. This statistic changes only during configuration updates. It does not take into account any addresses that may be reserved due to host reservation. The id is the the subnet-id of a given subnet. This statistic is exposed for each subnet separately, and is reset during a reconfiguration event.
subnet[id].pool[pid].total-addresses integer Total number of addresses available for DHCPv4 management for a given subnet pool; in other words, this is the count of all addresses in configured subnet pool. This statistic changes only during configuration updates. It does not take into account any addresses that may be reserved due to host reservation. The id is the subnet-id of a given subnet. The pid is the pool-id of a given pool. This statistic is exposed for each subnet pool separately, and is reset during a reconfiguration event.
cumulative-assigned-addresses integer Cumulative number of addresses that have been assigned since server startup. It is incremented each time an address is assigned and is not reset when the server is reconfigured.
subnet[id].cumulative-assigned-addresses integer Cumulative number of assigned addresses in a given subnet. It increases every time a new lease is allocated (as a result of receiving a DHCPREQUEST message) and never decreases. The id is the subnet-id of the subnet. This statistic is exposed for each subnet separately, and is reset during a reconfiguration event.
subnet[id].pool[pid].cumulative-assigned-addresses integer Cumulative number of assigned addresses in a given subnet pool. It increases every time a new lease is allocated (as a result of receiving a DHCPREQUEST message) and never decreases. The id is the subnet-id of the subnet. The pid is the pool-id of the pool. This statistic is exposed for each subnet pool separately, and is reset during a reconfiguration event.
subnet[id].assigned-addresses integer Number of assigned addresses in a given subnet. It increases every time a new lease is allocated (as a result of receiving a DHCPREQUEST message) and decreases every time a lease is released (a DHCPRELEASE message is received) or expires. The id is the subnet-id of the subnet. This statistic is exposed for each subnet separately, and is reset during a reconfiguration event.
subnet[id].pool[pid].assigned-addresses integer Number of assigned addresses in a given subnet pool. It increases every time a new lease is allocated (as a result of receiving a DHCPREQUEST message) and decreases every time a lease is released (a DHCPRELEASE message is received) or expires. The id is the subnet-id of the subnet. The pid is the pool-id of the pool. This statistic is exposed for each subnet pool separately, and is reset during a reconfiguration event.
reclaimed-leases integer Number of expired leases that have been reclaimed since server startup. It is incremented each time an expired lease is reclaimed and never decreases. It can be used as a long-term indicator of how many actual leases have been reclaimed. This is a global statistic that covers all subnets.
subnet[id].reclaimed-leases integer Number of expired leases associated with a given subnet that have been reclaimed since server startup. It is incremented each time an expired lease is reclaimed. The id is the subnet-id of a given subnet. This statistic is exposed for each subnet separately.
subnet[id].pool[pid].reclaimed-leases integer Number of expired leases associated with a given subnet pool that have been reclaimed since server startup. It is incremented each time an expired lease is reclaimed. The id is the subnet-id of a given subnet. The pid is the pool-id of the pool. This statistic is exposed for each subnet pool separately.
declined-addresses integer Number of IPv4 addresses that are currently declined; a count of the number of leases currently unavailable. Once a lease is recovered, this statistic is decreased; ideally, this statistic should be zero. If this statistic is non-zero or increasing, a network administrator should investigate whether there is a misbehaving device in the network. This is a global statistic that covers all subnets.
subnet[id].declined-addresses integer Number of IPv4 addresses that are currently declined in a given subnet; a count of the number of leases currently unavailable. Once a lease is recovered, this statistic is decreased; ideally, this statistic should be zero. If this statistic is non-zero or increasing, a network administrator should investigate whether there is a misbehaving device in the network. The id is the subnet-id of a given subnet. This statistic is exposed for each subnet separately.
subnet[id].pool[pid].declined-addresses integer Number of IPv4 addresses that are currently declined in a given subnet pool; a count of the number of leases currently unavailable. Once a lease is recovered, this statistic is decreased; ideally, this statistic should be zero. If this statistic is non-zero or increasing, a network administrator should investigate whether there is a misbehaving device in the network. The id is the subnet-id of a given subnet. The pid is the pool-id of the pool. This statistic is exposed for each subnet pool separately.
reclaimed-declined-addresses integer Number of IPv4 addresses that were declined, but have now been recovered. Unlike declined-addresses, this statistic never decreases. It can be used as a long-term indicator of how many actual valid declines were processed and recovered from. This is a global statistic that covers all subnets.
subnet[id].reclaimed-declined-addresses integer Number of IPv4 addresses that were declined, but have now been recovered. Unlike declined-addresses, this statistic never decreases. It can be used as a long-term indicator of how many actual valid declines were processed and recovered from. The id is the subnet-id of a given subnet. This statistic is exposed for each subnet separately.
subnet[id].pool[pid].reclaimed-declined-addresses integer Number of IPv4 addresses that were declined, but have now been recovered. Unlike declined-addresses, this statistic never decreases. It can be used as a long-term indicator of how many actual valid declines were processed and recovered from. The id is the subnet-id of a given subnet. The pid is the pool-id of the pool. This statistic is exposed for each subnet pool separately.
pkt4-lease-query-received integer Number of IPv4 DHCPLEASEQUERY packets received. (Only exists if Leasequery hook library is loaded.)
pkt4-lease-query-response-unknown-sent integer Number of IPv4 DHCPLEASEUNKNOWN responses sent. (Only exists if Leasequery hook library is loaded.)
pkt4-lease-query-response-unassigned-sent integer Number of IPv4 DHCPLEASEUNASSIGNED responses sent. (Only exists if Leasequery hook library is loaded.)
pkt4-lease-query-response-active-sent integer Number of IPv4 DHCPLEASEACTIVE responses sent. (Only exists if Leasequery hook library is loaded.)
v4-allocation-fail integer Number of total address allocation failures for a particular client. This consists in the number of lease allocation attempts that the server made before giving up and was unable to use any of the address pools. This is a global statistic that covers all subnets.
subnet[id].v4-allocation-fail integer Number of total address allocation failures for a particular client. This consists in the number of lease allocation attempts that the server made before giving up and was unable to use any of the address pools. The id is the subnet-id of a given subnet. This statistic is exposed for each subnet separately.
v4-allocation-fail-shared-network integer Number of address allocation failures for a particular client connected to a shared network. This is a global statistic that covers all subnets.
subnet[id].v4-allocation-fail-shared-network integer Number of address allocation failures for a particular client connected to a shared network. The id is the subnet-id of a given subnet. This statistic is exposed for each subnet separately.
v4-allocation-fail-subnet integer Number of address allocation failures for a particular client connected to a subnet that does not belong to a shared network. This is a global statistic that covers all subnets.
subnet[id].v4-allocation-fail-subnet integer Number of address allocation failures for a particular client connected to a subnet that does not belong to a shared network. The id is the subnet-id of a given subnet. This statistic is exposed for each subnet separately.
v4-allocation-fail-no-pools integer Number of address allocation failures because the server could not use any configured pools for a particular client. It is also possible that all of the subnets from which the server attempted to assign an address lack address pools. In this case, it should be considered misconfiguration if an operator expects that some clients should be assigned dynamic addresses. This is a global statistic that covers all subnets.
subnet[id].v4-allocation-fail-no-pools integer Number of address allocation failures because the server could not use any configured pools for a particular client. It is also possible that all of the subnets from which the server attempted to assign an address lack address pools. In this case, it should be considered misconfiguration if an operator expects that some clients should be assigned dynamic addresses. The id is the subnet-id of a given subnet. This statistic is exposed for each subnet separately.
v4-allocation-fail-classes integer Number of address allocation failures when the client's packet belongs to one or more classes. There may be several reasons why a lease was not assigned. One of them may be a case when all pools require packet to belong to certain classes and the incoming packet didn't belong to any of them. Another case where this information may be useful is to point out that the pool reserved to a given class has ran out of addresses. This is a global statistic that covers all subnets.
subnet[id].v4-allocation-fail-classes integer Number of address allocation failures when the client's packet belongs to one or more classes. There may be several reasons why a lease was not assigned. One of them may be a case when all pools require packet to belong to certain classes and the incoming packet didn't belong to any of them. Another case where this information may be useful is to point out that the pool reserved to a given class has ran out of addresses. The id is the subnet-id of a given subnet. This statistic is exposed for each subnet separately.
v4-lease-reuses integer Number of times an IPv4 lease had its CLTT increased in memory and its expiration time left unchanged in persistent storage as part of the lease caching feature. This is referred to as a lease reuse. This statistic is global.
subnet[id].v4-lease-reuses integer Number of times an IPv4 lease had its CLTT increased in memory and its expiration time left unchanged in persistent storage as part of the lease caching feature. This is referred to as a lease reuse. This statistic is on a per-subnet basis. The id is the subnet-id of a given subnet.
v4-reservation-conflicts integer Number of host reservation allocation conflicts which have occurred across every subnet. When a client sends a DHCP Discover and is matched to a host reservation which is already leased to another client, this counter is increased by 1.
subnet[id].v4-reservation-conflicts integer Number of host reservation allocation conflicts which have occurred in a specific subnet. When a client sends a DHCP Discover and is matched to a host reservation which is already leased to another client, this counter is increased by 1.

Note

The pool ID can be configured on each pool by explicitly setting the pool-id parameter in the pool parameter map. If not configured, pool-id defaults to 0. The statistics related to pool ID 0 refer to all the statistics of all the pools that have unconfigured pool-id. The pool ID does not need to be unique within the subnet or across subnets. The statistics regarding a specific pool ID within a subnet will be combined with the other statistics of all other pools with the same pool ID in the respective subnet.

Note

This section describes DHCPv4-specific statistics. For a general overview and usage of statistics, see Statistics.

The DHCPv4 server provides two global parameters to control the default sample limits of statistics:

  • statistic-default-sample-count - determines the default maximum number of samples which are kept. The special value of 0 indicates that a default maximum age should be used.
  • statistic-default-sample-age - determines the default maximum age in seconds of samples which are kept.

For instance, to reduce the statistic-keeping overhead, set the default maximum sample count to 1 so only one sample is kept:

"Dhcp4": {
    "statistic-default-sample-count": 1,
    "subnet4": [
        {
            ...
        },
        ...
    ],
    ...
}

Statistics can be retrieved periodically to gain more insight into Kea operations. One tool that leverages that capability is ISC Stork. See Monitoring Kea With Stork for details.

8.9. Management API for the DHCPv4 Server

The management API allows the issuing of specific management commands, such as statistics retrieval, reconfiguration, or shutdown. For more details, see Management API. Currently, the only supported communication channel type is the UNIX stream socket. By default there are no sockets open; to instruct Kea to open a socket, the following entry in the configuration file can be used:

"Dhcp4": {
    "control-socket": {
        "socket-type": "unix",
        "socket-name": "/path/to/the/unix/socket"
    },

    "subnet4": [
        {
            ...
        },
        ...
    ],
    ...
}

The length of the path specified by the socket-name parameter is restricted by the maximum length for the UNIX socket name on the administrator's operating system, i.e. the size of the sun_path field in the sockaddr_un structure, decreased by 1. This value varies on different operating systems, between 91 and 107 characters. Typical values are 107 on Linux and 103 on FreeBSD.

Communication over the control channel is conducted using JSON structures. See the Control Channel section in the Kea Developer's Guide for more details.

The DHCPv4 server supports the following operational commands:

as described in Commands Supported by Both the DHCPv4 and DHCPv6 Servers. In addition, it supports the following statistics-related commands:

as described in Commands for Manipulating Statistics.

8.10. User Contexts in IPv4

Kea allows the loading of hook libraries that can sometimes benefit from additional parameters. If such a parameter is specific to the whole library, it is typically defined as a parameter for the hook library. However, sometimes there is a need to specify parameters that are different for each pool.

See Comments and User Context for additional background regarding the user-context idea. See User Contexts in Hooks for a discussion from the hooks perspective.

User contexts can be specified at global scope; at the shared-network, subnet, pool, client-class, option-data, or definition level; and via host reservation. One other useful feature is the ability to store comments or descriptions.

Let's consider an imaginary case of devices that have colored LED lights. Depending on their location, they should glow red, blue, or green. It would be easy to write a hook library that would send specific values, maybe as a vendor option. However, the server has to have some way to specify that value for each pool. This need is addressed by user contexts. In essence, any user data can be specified in the user context as long as it is a valid JSON map. For example, the aforementioned case of LED devices could be configured in the following way:

"Dhcp4": {
    "subnet4": [
        {
        "id": 1,
        "subnet": "192.0.2.0/24",
        "pools": [
        {
            "pool": "192.0.2.10 - 192.0.2.20",
            # This is pool specific user context
            "user-context": { "color": "red" }
        }
        ],

        # This is a subnet-specific user context. Any type
        # of information can be entered here as long as it is valid JSON.
        "user-context": {
            "comment": "network on the second floor",
            "last-modified": "2017-09-04 13:32",
            "description": "you can put anything you like here",
            "phones": [ "x1234", "x2345" ],
            "devices-registered": 42,
            "billing": false
        }
    }
    ]
}

Kea does not interpret or use the user-context information; it simply stores it and makes it available to the hook libraries. It is up to each hook library to extract that information and use it. The parser translates a comment entry into a user context with the entry, which allows a comment to be attached inside the configuration itself.

8.11. Supported DHCP Standards

The following standards are currently supported in Kea:

  • BOOTP Vendor Information Extensions, RFC 1497: This requires the open source BOOTP hook to be loaded. See libdhcp_bootp.so: Support for BOOTP Clients for details.
  • Dynamic Host Configuration Protocol, RFC 2131: Supported messages are DHCPDISCOVER (1), DHCPOFFER (2), DHCPREQUEST (3), DHCPRELEASE (7), DHCPINFORM (8), DHCPACK (5), and DHCPNAK(6).
  • DHCP Options and BOOTP Vendor Extensions, RFC 2132: Supported options are PAD (0), END(255), Message Type(53), DHCP Server Identifier (54), Domain Name (15), DNS Servers (6), IP Address Lease Time (51), Subnet Mask (1), and Routers (3).
  • The IPv4 Subnet Selection Option for DHCP, RFC 3011: The subnet-selection option is supported; if received in a packet, it is used in the subnet-selection process.
  • DHCP Relay Agent Information Option, RFC 3046: Relay Agent Information, Circuit ID, and Remote ID options are supported.
  • Link Selection sub-option for the Relay Agent Option, RFC 3527: The link selection sub-option is supported.
  • Vendor-Identifying Vendor Options for Dynamic Host Configuration Protocol version 4, RFC 3925: The Vendor-Identifying Vendor Class and Vendor-Identifying Vendor-Specific Information options are supported.
  • Subscriber-ID Suboption for the DHCP Relay Agent Option, RFC 3993: The Subscriber-ID option is supported.
  • The Dynamic Host Configuration Protocol (DHCP) Client Fully Qualified Domain Name (FQDN) Option, RFC 4702: The Kea server is able to handle the Client FQDN option. Also, it is able to use the kea-dhcp-ddns component to initiate appropriate DNS Update operations.
  • Resolution of Fully Qualified Domain Name (FQDN) Conflicts among Dynamic Host Configuration Protocol (DHCP) Clients, RFC 4703: The DHCPv6 server uses a DHCP-DDNS server to resolve conflicts.
  • Client Identifier Option in DHCP Server Replies, RFC 6842: The server by default sends back the client-id option. That capability can be disabled. See Echoing Client-ID (RFC 6842) for details.
  • Generalized UDP Source Port for the DHCP Relay Agent Option, RFC 8357: The Kea server handles the Relay Agent Information Source Port sub-option in a received message, remembers the UDP port, and sends back a reply to the same relay agent using this UDP port.
  • Captive-Portal Identification in DHCP and Router Advertisements (RAs), RFC 8910: The Kea server can configure both v4 and v6 versions of the captive portal options.
  • IPv6-Only Preferred Option for DHCPv4, RFC 8925: The Kea server is able to designate its pools and subnets as IPv6-Only Preferred and send back the v6-only-preferred option to clients that requested it.
  • Server Identifier Override sub-option for the Relay Agent Option, RFC 5107: The server identifier override sub-option is supported. The implementation is not complete according to the RFC, because the server does not store the RAI, but the functionality handles expected use cases.

8.11.1. Known RFC Violations

In principle, Kea aspires to be a reference implementation and aims to implement 100% of the RFC standards. However, in some cases there are practical aspects that prevent Kea from completely adhering to the text of the RFC documents.

  • RFC 2131, page 30, says that if the incoming DHCPREQUEST packet has no "requested IP address" option and ciaddr is not set, the server is supposed to respond with NAK. However, broken clients exist that will always send a DHCPREQUEST without those options indicated. In that event, Kea accepts the DHCPREQUEST, assigns an address, and responds with an ACK.
  • RFC 2131, table 5, says that messages of type DHCPDECLINE or DHCPRELEASE must have the server identifier set and should be dropped if that option is missing. However, ISC DHCP does not enforce this, presumably as a compatibility effort for broken clients, and the Kea team decided to follow suit.

8.12. DHCPv4 Server Limitations

These are the current known limitations of the Kea DHCPv4 server software. Most of them are reflections of the current stage of development and should be treated as “not implemented yet”, rather than as actual limitations. However, some of them are implications of the design choices made. Those are clearly marked as such.

  • On the Linux and BSD system families, DHCP messages are sent and received over raw sockets (using LPF and BPF) and all packet headers (including data link layer, IP, and UDP headers) are created and parsed by Kea, rather than by the system kernel. Currently, Kea can only parse the data-link layer headers with a format adhering to the IEEE 802.3 standard, and assumes this data-link-layer header format for all interfaces. Thus, Kea does not work on interfaces which use different data-link-layer header formats (e.g. Infiniband).
  • The DHCPv4 server does not verify that an assigned address is unused. According to RFC 2131, the allocating server should verify that an address is not used by sending an ICMP echo request.

8.13. Kea DHCPv4 Server Examples

A collection of simple-to-use examples for the DHCPv4 component of Kea is available with the source files, located in the doc/examples/kea4 directory.

8.14. Configuration Backend in DHCPv4

In the Kea Configuration Backend section we have described the Configuration Backend (CB) feature, its applicability, and its limitations. This section focuses on the usage of the CB with the Kea DHCPv4 server. It lists the supported parameters, describes limitations, and gives examples of DHCPv4 server configurations to take advantage of the CB. Please also refer to the corresponding section Configuration Backend in DHCPv6 for DHCPv6-specific usage of the CB.

8.14.1. Supported Parameters

The ultimate goal for the CB is to serve as a central configuration repository for one or multiple Kea servers connected to a database. In currently supported Kea versions, only a subset of the DHCPv4 server parameters can be configured in the database. All other parameters must be specified in the JSON configuration file, if required.

All supported parameters can be configured via libdhcp_cb_cmds.so. The general rule is that scalar global parameters are set using remote-global-parameter4-set; shared-network-specific parameters are set using remote-network4-set; and subnet-level and pool-level parameters are set using remote-subnet4-set. Whenever there is an exception to this general rule, it is highlighted in the table. Non-scalar global parameters have dedicated commands; for example, the global DHCPv4 options (option-data) are modified using remote-option4-global-set. Client classes, together with class-specific option definitions and DHCPv4 options, are configured using the remote-class4-set command.

The Configuration Sharing and Server Tags section explains the concept of shareable and non-shareable configuration elements and the limitations for sharing them between multiple servers. In the DHCP configuration (both DHCPv4 and DHCPv6), the shareable configuration elements are subnets and shared networks. Thus, they can be explicitly associated with multiple server tags. The global parameters, option definitions, and global options are non-shareable and can be associated with only one server tag. This rule does not apply to the configuration elements associated with all servers. Any configuration element associated with all servers (using the all keyword as a server tag) is used by all servers connecting to the configuration database.

The following table lists DHCPv4-specific parameters supported by the Configuration Backend, with an indication of the level of the hierarchy at which it is currently supported.

List of DHCPv4 parameters supported by the Configuration Backend
Parameter Global Client Class Shared Network Subnet Pool
4o6-interface n/a n/a n/a yes n/a
4o6-interface-id n/a n/a n/a yes n/a
4o6-subnet n/a n/a n/a yes n/a
allocator yes n/a yes yes n/a
boot-file-name yes yes yes yes n/a
cache-max-age yes n/a no no n/a
cache-threshold yes n/a no no n/a
calculate-tee-times yes n/a yes yes n/a
client-class n/a n/a yes yes yes
ddns-send-update yes n/a yes yes n/a
ddns-override-no-update yes n/a yes yes n/a
ddns-override-client-update yes n/a yes yes n/a
ddns-replace-client-name yes n/a yes yes n/a
ddns-generated-prefix yes n/a yes yes n/a
ddns-qualifying-suffix yes n/a yes yes n/a
decline-probation-period yes n/a n/a n/a n/a
dhcp4o6-port yes n/a n/a n/a n/a
echo-client-id yes n/a n/a n/a n/a
hostname-char-set no n/a no no n/a
hostname-char-replacement no n/a no no n/a
interface n/a n/a yes yes n/a
match-client-id yes n/a yes yes n/a
min-valid-lifetime yes yes yes yes n/a
max-valid-lifetime yes yes yes yes n/a
next-server yes yes yes yes n/a
option-data yes (via remote-option4-global-set) yes yes yes yes
option-def yes (via remote-option-def4-set) yes n/a n/a n/a
rebind-timer yes n/a yes yes n/a
renew-timer yes n/a yes yes n/a
server-hostname yes yes yes yes n/a
valid-lifetime yes yes yes yes n/a
relay n/a n/a yes yes n/a
require-client-classes no n/a yes yes yes
reservation-mode yes n/a yes yes n/a
reservations-global yes n/a yes yes n/a
reservations-in-subnet yes n/a yes yes n/a
reservations-out-of-pool yes n/a yes yes n/a
t1-percent yes n/a yes yes n/a
t2-percent yes n/a yes yes n/a
  • yes - indicates that the parameter is supported at the given level of the hierarchy and can be configured via the Configuration Backend.
  • no - indicates that a parameter is supported at the given level of the hierarchy but cannot be configured via the Configuration Backend.
  • n/a - indicates that a given parameter is not applicable at the particular level of the hierarchy or that the server does not support the parameter at that level.

8.14.2. Enabling the Configuration Backend

Consider the following configuration snippet, which uses a MySQL configuration database:

{
  "Dhcp4": {
    "server-tag": "my DHCPv4 server",
    "config-control": {
        "config-databases": [
        {
            "type": "mysql",
            "name": "kea",
            "user": "kea",
            "password": "kea",
            "host": "192.0.2.1",
            "port": 3302
        }
        ],
        "config-fetch-wait-time": 20
    },
    "hooks-libraries": [
    {
        "library": "/usr/local/lib/kea/hooks/libdhcp_mysql_cb.so"
    }, {
        "library": "/usr/local/lib/kea/hooks/libdhcp_cb_cmds.so"
    }
    ]
  }
}

The config-control map contains two parameters. config-databases is a list that contains one element, which includes the database type, its location, and the credentials to be used to connect to this database. (Note that the parameters specified here correspond to the database specification for the lease database backend and hosts database backend.) Currently only one database connection can be specified on the config-databases list. The server connects to this database during startup or reconfiguration, and fetches the configuration available for this server from the database. This configuration is merged into the configuration read from the configuration file.

The following snippet illustrates the use of a PostgreSQL database:

{
  "Dhcp4": {
    "server-tag": "my DHCPv4 server",
    "config-control": {
        "config-databases": [
        {
            "type": "postgresql",
            "name": "kea",
            "user": "kea",
            "password": "kea",
            "host": "192.0.2.1",
            "port": 5432
        }
        ],
        "config-fetch-wait-time": 20
    },
    "hooks-libraries": [
    {
        "library": "/usr/local/lib/kea/hooks/libdhcp_pgsql_cb.so"
    }, {
        "library": "/usr/local/lib/kea/hooks/libdhcp_cb_cmds.so"
    }
    ]
  }
}

Note

Whenever there is a conflict between the parameters specified in the configuration file and the database, the parameters from the database take precedence. We strongly recommend avoiding the duplication of parameters in the file and the database, but this recommendation is not enforced by the Kea servers. In particular, if the subnets' configuration is sourced from the database, we recommend that all subnets be specified in the database and that no subnets be specified in the configuration file. It is possible to specify the subnets in both places, but the subnets in the configuration file with overlapping IDs and/or prefixes with the subnets from the database will be superseded by those from the database.

Once the Kea server is configured, it starts periodically polling the database for configuration changes. The polling frequency is controlled by the config-fetch-wait-time parameter, expressed in seconds; it is the period between the time when the server completed its last poll (and possibly the local configuration update) and the time when it will begin polling again. In the example above, this period is set to 20 seconds. This means that after adding a new configuration into the database (e.g. adding a new subnet), it will take up to 20 seconds (plus the time needed to fetch and apply the new configuration) before the server starts using this subnet. The lower the config-fetch-wait-time value, the shorter the time for the server to react to incremental configuration updates in the database. On the other hand, polling the database too frequently may impact the DHCP server's performance, because the server needs to make at least one query to the database to discover any pending configuration updates. The default value of config-fetch-wait-time is 30 seconds.

The config-backend-pull command can be used to force the server to immediately poll any configuration changes from the database and avoid waiting for the next fetch cycle.

In the configuration examples above, two hook libraries are loaded. The first is a library which implements the Configuration Backend for a specific database type: libdhcp_mysql_cb.so provides support for MySQL and libdhcp_pgsql_cb.so provides support for PostgreSQL. The library loaded must match the database type specified within the config-control parameter or an will error be logged when the server attempts to load its configuration and the load will fail.

The second hook library, libdhcp_cb_cmds.so, is optional. It should be loaded when the Kea server instance is to be used to manage the configuration in the database. See the libdhcp_cb_cmds.so: Configuration Backend Commands section for details. This hook library is only available to ISC customers with a paid support contract.

8.15. Kea DHCPv4 Compatibility Configuration Parameters

ISC's intention is for Kea to follow the RFC documents to promote better standards compliance. However, many buggy DHCP implementations already exist that cannot be easily fixed or upgraded. Therefore, Kea provides an easy-to-use compatibility mode for broken or non-compliant clients. For that purpose, the compatibility option must be enabled to permit uncommon practices:

{
  "Dhcp4": {
    "compatibility": {
    }
  }
}

8.15.1. Lenient Option Parsing

By default, tuple fields defined in custom options are parsed as a set of length-value pairs.

With "lenient-option-parsing": true, if a length ever exceeds the rest of the option's buffer, previous versions of Kea returned a log message unable to parse the opaque data tuple, the buffer length is x, but the tuple length is y with x < y; this no longer occurs. Instead, the value is considered to be the rest of the buffer, or in terms of the log message above, the tuple length y becomes x.

{
  "Dhcp4": {
    "compatibility": {
      "lenient-option-parsing": true
    }
  }
}

8.15.2. Ignore DHCP Server Identifier

With "ignore-dhcp-server-identifier": true, the server does not check the address in the DHCP Server Identifier option, i.e. whether a query is sent to this server or another one (and in the second case dropping the query).

{
  "Dhcp4": {
    "compatibility": {
      "ignore-dhcp-server-identifier": true
    }
  }
}

8.15.4. Exclude First Last Addresses in /24 Subnets or Larger

The exclude-first-last-24 compatibility flag is described in Configuration of IPv4 Address Pools (when true .0 and .255 addresses are excluded from subnets with prefix length less than or equal to 24).

8.16. Address Allocation Strategies in DHCPv4

A DHCP server follows a complicated algorithm to select an IPv4 address for a client. It prefers assigning specific addresses requested by the client and the addresses for which the client has reservations.

If the client requests no particular address and has no reservations, or other clients are already using any requested addresses, the server must find another available address within the configured pools. A server function called an "allocator" is responsible in Kea for finding an available address in such a case.

The Kea DHCPv4 server provides configuration parameters to select different allocators at the global, shared-network, and subnet levels. Consider the following example:

{
    "Dhcp4": {
        "allocator": "random",
        "subnet4": [
            {
                "id": 1,
                "subnet": "10.0.0.0/8",
                "allocator": "iterative"
            },
            {
                "id": 2,
                "subnet": "192.0.2.0/24"
            }
        ]
    }
}

This allocator overrides the default iterative allocation strategy at the global level and selects the random allocation instead. The random allocation will be used for the subnet with ID 2, while the iterative allocation will be used for the subnet with ID 1.

The following sections describe the supported allocators and their recommended uses.

8.16.1. Allocators Comparison

In the table below, we briefly compare the supported allocators. The detailed allocators' descriptions are in later sections.

Comparison of the lease allocators supported by Kea DHCPv4
Allocator Low Utilization Performance High Utilization Performance Lease Randomization Startup/Configuration Memory Usage
Iterative very high low no very fast low
Random high low yes very fast high (varying)
Free Lease Queue high high yes slow (depends on pool sizes) high (varying)

8.16.2. Iterative Allocator

This is the default allocator used by the Kea DHCPv4 server. It remembers the last offered address and offers this address, increased by one, to the next client. For example, it may offer addresses in this order: 192.0.2.10, 192.0.2.11, 192.0.2.12, and so on. The time to find and offer the next address is very short; thus, this is the most performant allocator when pool utilization is low and there is a high probability that the next address is available.

The iterative allocation underperforms when multiple DHCP servers share a lease database or are connected to a cluster. The servers tend to offer and allocate the same blocks of addresses to different clients independently, which causes many allocation conflicts between the servers and retransmissions by clients. A random allocation addresses this issue by dispersing the allocation order.

8.16.3. Random Allocator

The random allocator uses a uniform randomization function to select offered addresses from subnet pools. It is suitable in deployments where multiple servers are connected to a shared database or a database cluster. By dispersing the offered addresses, the servers minimize the risk of allocating the same address to two different clients at the same or nearly the same time. In addition, it improves the server's resilience against attacks based on allocation predictability.

The random allocator is, however, slightly slower than the iterative allocator. Moreover, it increases the server's memory consumption because it must remember randomized addresses to avoid offering them repeatedly. Memory consumption grows with the number of offered addresses; in other words, larger pools and more clients increase memory consumption by random allocation.

The following configuration snippet shows how to select the random allocator for a subnet:

{
    "Dhcp4": {
        "allocator": "random",
        "subnet4": [
            {
                "id": 1,
                "subnet": "10.0.0.0/8",
                "allocator": "random"
            }
        ]
    }
}

8.16.4. Free Lease Queue Allocator

This is a sophisticated allocator whose use should be considered in subnets with highly utilized address pools. In such cases, it can take a considerable amount of time for the iterative or random allocator to find an available address, because they must repeatedly check whether there is a valid lease for an address they will offer. The number of checks can be as high as the number of addresses in the subnet when the subnet pools are exhausted, which can have a direct negative impact on the DHCP response time for each request.

The Free Lease Queue (FLQ) allocator tracks lease allocations and de-allocations and maintains a running list of available addresses for each address pool. It allows an available lease to be selected within a constant time, regardless of the subnet pools' utilization. The allocator continuously updates the list of free leases by removing any allocated leases and adding released or reclaimed ones.

The following configuration snippet shows how to select the FLQ allocator for a subnet:

{
    "Dhcp4": {
        "subnet4": [
            {
                "id": 1,
                "subnet": "192.0.2.0/24",
                "allocator": "flq"
            }
        ]
    }
}

There are several considerations that the administrator should take into account before using this allocator. The FLQ allocator can heavily impact the server's startup and reconfiguration time, because the allocator has to populate the list of free leases for each subnet where it is used. These delays can be observed both during the configuration reload and when the subnets are created using libdhcp_subnet_cmds.so. The allocator increases the memory consumption to hold the list of free leases, proportional to the total size of the address pools for which this allocator is used. Finally, lease reclamation must be enabled with a low value of the reclaim-timer-wait-time parameter, to ensure that the server frequently collects expired leases and makes them available for allocation via the free lease queue. Expired leases are not considered free by the allocator until they are reclaimed by the server. See Lease Reclamation for more details about the lease reclamation process.

We recommend that the FLQ allocator be selected only after careful consideration. For example, using it for a subnet with a /8 pool may delay the server's startup by 15 seconds or more. On the other hand, the startup delay and the memory consumption increase should be acceptable for subnets with a /16 pool or smaller. We also recommend specifying another allocator type in the global configuration settings and overriding this selection at the subnet or shared-network level, to use the FLQ allocator only for selected subnets. That way, when a new subnet is added without an allocator specification, the global setting is used, thus avoiding unnecessary impact on the server's startup time.

Like the random allocator, the FLQ allocator offers leases in random order, which makes it suitable for use with a shared lease database.