]>

It is recommended that the Kea DHCPv4 server be started and stopped using keactrl (described in ). However, it is also possible to run the server directly: it 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 and "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 port - specifies UDP port on which the server will listen. This is only useful during testing, as a DHCPv4 server listening on ports other than the standard ones will not be able to handle regular DHCPv4 queries. -t file - specifies the configuration file to be tested. Kea-dhcp4 will attempt to load it, and will conduct sanity checks. Note that certain checks are possible only while running the actual server. The actual status is reported with exit code (0 = configuration looks ok, 1 = error encountered). Kea will print out log messages to standard output and error to standard error when testing configuration. -v - prints out the Kea version and exits. -V - prints out the Kea extended version with additional parameters and exits. The listing includes the versions of the libraries dynamically linked to Kea. -W - prints out 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 config.report may also be accessed more directly. The following command may be used to extract this information. The binary path may be found in the install directory or in the .libs subdirectory in the source tree. For example kea/src/bin/dhcp4/.libs/kea-dhcp4. strings path/kea-dhcp4 | sed -n 's/;;;; //p' On start-up, the server will detect available network interfaces and will attempt to open UDP sockets on all interfaces mentioned in the configuration file. Since the DHCPv4 server opens privileged ports, it requires root access. Make sure you run this daemon as root. During startup the server will attempt to create a PID file of the form: localstatedir]/[conf name].kea-dhcp6.pid where: localstatedir: The value as passed into the build configure script. It defaults to "/usr/local/var". (Note that this value may be overridden at run time by setting the environment variable KEA_PIDFILE_DIR. This is intended primarily for testing purposes.) conf name: The configuration file name used to start the server, minus all preceding path and 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 will issue a DHCP4_ALREADY_RUNNING log message and exit. 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. It detects the key combination and shuts down gracefully.

This section explains how to configure the DHCPv4 server using the Kea configuration backend. (Kea configuration using any other backends is outside of scope of this document.) Before DHCPv4 is started, its configuration file has to be created. The basic configuration is as follows: { # DHCPv4 configuration starts in this line "Dhcp4": { # First we set up global values "valid-lifetime": 4000, "renew-timer": 1000, "rebind-timer": 2000, # Next we setup 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/kea/dhcp4.leases" }, # Finally, we list the subnets from which we will be leasing addresses. "subnet4": [ { "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 together 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 consists of one or more objects. In this specific example, we have only one object, called Dhcp4. This is a simplified configuration, as usually there will be additional objects, like Logging or DhcpDdns, but we omit them now for clarity. 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 the general case, the order in which those parameters appear does not matter. There are two caveats here though. The first one is to remember that the configuration file must be well formed JSON. That means 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, keep in mind that 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 your configuration file. Moving onto the DHCPv4 configuration elements, the first few elements define some global parameters. valid-lifetime defines for how long the addresses (leases) given out by the server are valid. If nothing changes, a client that got an address is allowed to use it 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 T1 and T2 timers that govern when the client will begin the renewal and rebind procedures. Note that renew-timer and rebind-timer are optional. If they are not specified the client will select values for T1 and T2 timers according to the RFC 2131. The interfaces-config map specifies the server configuration concerning the network interfaces, on which the server should listen to the 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. Had we wanted to listen on two interfaces, the interfaces-config would look like this: "interfaces-config": { "interfaces": [ "eth0", "eth1" ] }, The next couple of 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 file. This is a very simple configuration. 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. Had there been 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 has to have at least two parameters: subnet (that 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. Had more than one been 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": [ { "pools": [ { "pool": "192.0.2.1 - 192.0.2.200" } ], "subnet": "192.0.2.0/24" }, { "pools": [ { "pool": "192.0.3.100 - 192.0.3.200" } ], "subnet": "192.0.3.0/24" }, { "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 in may be preferable to use more compact notation. After all the parameters have been specified, we have two contexts open: global and Dhcp4, hence we need two closing curly brackets to close them. In a real life configuration file there most likely would be additional components defined such as Logging or DhcpDdns, so the closing brace would be followed by a comma and another object definition.

All leases issued by the server are stored in the lease database. Currently there are four database backends available: memfile (which is the default backend), MySQL, PostgreSQL and Cassandra.

The server is able to store lease data in different repositories. Larger deployments may elect to store leases in a database. describes this option. In typical smaller deployments though, the server will store 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 file backend (Memfile) is controlled through the Dhcp4/lease-database parameters. The type parameter is mandatory and it specifies which storage for leases the server should use. The value of "memfile" indicates that the file should be used as the storage. 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 is set to true at all times, during the server's normal operation. Not writing leases to disk will mean that if a server is restarted (e.g. after a power failure), it will not know what addresses have been assigned. As a result, it may hand out addresses to new clients 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 will be recorded. The default value for this parameter is "[kea-install-dir]/var/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 detailed fashion further in this section. The default value of the lfc-interval is 3600. A value of 0 disables the LFC. 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 } } This configuration selects the /tmp/kea-leases4.csv as the storage for lease information and enables persistence (writing lease updates to this file). It also configures the backend perform the periodic cleanup of the lease files, executed every 30 minutes. 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 the 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 it 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 the server startup, it assumes that the latest lease entry for the client is the valid one. The previous entries are discarded. This means that the server can re-construct the 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 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 new cleanup is started while the previous cleanup instance is still running, if the lfc-interval is too short. The server would recover from this by skipping the new cleanup when it detects that the previous cleanup is still in progress. But it implies that the actual cleanups will be triggered more rarely than configured. Moreover, triggering a new cleanup adds an overhead to the server which will not be 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 is selected in a way that would allow for the LFC to complete the cleanup before a new cleanup is triggered. Lease file cleanup is performed by a separate process (in background) to avoid a performance impact on the server process. In order to avoid the conflicts between two processes both using the same lease files, the LFC process operates on the copy of the original lease file, rather than on the lease file used by the server to record lease updates. There are also other files being created as a side effect of the lease file cleanup. The detailed description of the LFC is located on the Kea wiki: .

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. Lease database configuration is controlled through the Dhcp4/lease-database parameters. The type of the database must be set to "memfile", "mysql", "postgresql" or "cql", 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 , or ). "Dhcp4": { "lease-database": { "name": "database-name" , ... }, ... } For Cassandra: "Dhcp4": { "lease-database": { "keyspace": "database-name" , ... }, ... } If the database is located on a different system to the DHCPv4 server, the database host name must also be specified. (It should be noted that this configuration may have a severe impact on server performance.): "Dhcp4": { "lease-database": { "host": "remote-host-name", ... }, ... } For Cassandra, multiple contact points can be provided: "Dhcp4": { "lease-database": { "contact-points": "remote-host-name[, ...] ", ... }, ... } The usual state of affairs will be to have the database on the same machine as the DHCPv4 server. In this case, set the value to the empty string: "Dhcp4": { "lease-database": { "host" : "", ... }, ... } For Cassandra: "Dhcp4": { "lease-database": { "contact-points": "", ... }, ... } Should the database use a port different than default, it may be specified as well: "Dhcp4": { "lease-database": { "port" : 12345, ... }, ... } Should the database be located on a different system, you 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. Note that host parameter is used by MySQL and PostgreSQL backends. Cassandra has a concept of contact points that could be used to contact the cluster, instead of a single IP or hostname. It takes a list of comma separated IP addresses. This may be specified as: "Dhcp4": { "lease-database": { "contact-points" : "192.0.2.1,192.0.2.2", ... }, ... } 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 also the default.)

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, a Kea server opens independent connections for each purpose, be it lease or hosts information. This arrangement gives the most flexibility. Kea can be used to keep leases and host reservations separately, but can also point to the same database. Currently the supported hosts database types are MySQL, PostgreSQL and Cassandra. Please note that usage of hosts storage is optional. A user can define all host reservations in the configuration file. That is the recommended way if the number of reservations is small. However, when the number of reservations grows it's more convenient to use host storage. Please note that both storage methods (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.

Hosts database configuration is controlled through the Dhcp4/hosts-database parameters. If enabled, the type of the database must be set to "mysql" or "postgresql". Other hosts backends may be added in later versions of Kea. "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 for instructions how to setup 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. (Again it should be noted that this configuration may have a severe impact on server performance.): "Dhcp4": { "hosts-database": { "host": remote-host-name, ... }, ... } The usual state of affairs will be to have the database 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 default, it may be specified as well: "Dhcp4": { "hosts-database": { "port" : 12345, ... }, ... } 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 also the default.)

In some deployments the database 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 inventory databases deployed, which contain substantially more information about the hosts than static reservations assigned to them. The inventory database can be used to create a view of a Kea hosts database and such 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 database 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 assign 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 would configure the database backend to operate in "read-write" mode, which is also a default configuration if the parameter is not specified. The readonly parameter is currently only supported for MySQL and PostgreSQL databases.

The DHCPv4 server has to 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 wildcard interface name (asterisk) concurrently with explicit interface names: "Dhcp4": { "interfaces-config": { "interfaces": [ "eth1", "eth3", "*" ] }, ... } It is anticipated that this form of usage will 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 the 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 don't have an address configured. This requires that the server injects the hardware address of the destination into the data link layer of the packet being sent to the client. The DHCPv4 server utilizes the raw sockets to achieve this, and builds the entire IP/UDP stack for the outgoing packets. The down side of raw socket use, however, is that incoming and outgoing packets bypass the firewalls (e.g. iptables). It is also troublesome to handle traffic on multiple IPv4 addresses assigned to the same interface, as raw sockets are bound to the interface and advanced packet filtering techniques (e.g. using the BPF) have to be used to receive unicast traffic on the desired addresses assigned to the interface, rather than capturing whole traffic reaching the interface to which the raw socket is bound. Therefore, in the deployments where the server doesn't have to provision the directly connected clients and only receives the unicast packets from the relay agents, the DHCP server should be configured to utilize IP/UDP datagram 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 specifies that the IP/UDP sockets will be opened on all interfaces on which the server listens, i.e. "eth1" and "eth3" in our case. If the 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 the UDP sockets can be used for relayed traffic only. When using the raw sockets, both the traffic from the directly connected clients and the relayed traffic will be 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. Hence, 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 interface, the "interface-name/address" notation should be used. Specifying the value raw as the socket type, doesn't guarantee that the raw sockets will be used! The use of raw sockets to handle the traffic from the directly connected clients is currently supported on Linux and BSD systems only. If the raw sockets are not supported on the particular OS, the server will issue a warning and fall back to use IP/UDP sockets. In typical environment the DHCP server is expected to send back a response on the same network interface on which the query is received. This is the default behavior. However, in some deployments it is desired that the outbound (response) packets will be sent as regular traffic and the outbound interface will be determined by the routing tables. This kind of asymetric traffic is uncommon, but valid. Kea now 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 inteface the query packet is received. This is the default behavior. The second one, use-routing tells Kea to send regular UDP packets and let the kernel's routing table to determine 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 re-detect value to false, for instance: "Dhcp4": { "interfaces-config": { "interfaces": [ "eth1", "eth3" ], "re-detect": false }, ... } Note interfaces are not re-detected during config-test.

The use of UDP sockets has certain benefits in deployments where the server receives only relayed traffic; these benefits are mentioned in . From the administrator's perspective it is often desirable to configure the system's firewall to filter out the 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 are focusing 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 will first use 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 the 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.

The subnet identifier is a unique number associated with a particular subnet. In principle, it is used to associate clients' leases with their respective subnets. When a subnet identifier is not specified for a subnet being configured, it will be automatically assigned by the configuration mechanism. The identifiers are assigned from 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 planned to implement a mechanism to preserve auto-generated subnet ids in a future version of Kea. However, the only remedy for this issue at present is to manually specify a unique identifier for each subnet. The following configuration will assign the specified subnet identifier to the 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.

The main role of a DHCPv4 server is address assignment. For this, the server has to be configured with at least one subnet and one pool of dynamic addresses for it to manage. 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 has decided that addresses from range 192.0.2.10 to 192.0.2.20 are going to be managed by the Dhcp4 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. Additional parameters will be added in future releases of Kea. 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 be also be managed by the server. It could be written as 192.0.2.64 to 192.0.2.127. Alternatively, 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, one could define the following pools: "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: "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 will also be able to allocate the first (typically network address) and the last (typically broadcast address) address from that pool. In the aforementioned example of pool 192.0.3.0/24, both 192.0.3.0 and 192.0.3.255 addresses may be assigned as well. This may be invalid in some network configurations. If you want to avoid this, please use the "min-max" notation.

One of the major features of the DHCPv4 server is 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" is returned to the client if the client has included this option in its message to the server. 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 some 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 only one of name or code is required, you don't need to specify both. Space has a default value of "dhcp4", so you can skip this as well if you define a regular (not encapsulated) DHCPv4 option. Finally, csv-format defaults to true, so it too can be skipped, unless you want to specify the option value as hexstring. 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 response when the client requests them at 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) so in: "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" }, ... ], ... }, ... ], ... } The Domain Name Servers option is always added to responses (the always-send is "sticky") but the value is the subnet one when the client is localized in the subnet. The name parameter specifies the option name. For a list of currently supported names, see 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 . 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. Data 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" }, ... ], ... } Care should be taken to use proper encoding when using hexadecimal format as Kea's ability to validate data correctness in hexadecimal is limited. Most of the parameters in the "option-data" structure are optional and can be omitted in some circumstances as discussed in the . It is possible to specify or override options on a per-subnet basis. If clients connected to most of your subnets are expected to get the same values of a given option, you should use global options: you can then override specific values for a small number of subnets. On the other hand, if you use different values in each subnet, it does not make sense to specify global option values (Dhcp4/option-data), rather you should set only subnet-specific values (Dhcp4/subnet[X]/option-data[Y]). 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. The server's administrator must not try to prioritize assignment of pool specific options by trying to order pools declarations in the server configuration. Future Kea releases may change the order in which options are assigned from the pools without any notice. The following configuration snippet demonstrates how to specify the DNS servers option, which will be 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" }, ... ], ... }, ... ], ... }, ... ], ... } The currently supported standard DHCPv4 options are listed in . The "Name" and "Code" are the values that should be used as a name in the option-data structures. "Type" designates the format of the data: the meanings of the various types is given in . When a data field is a string, and that string contains the comma (,; U+002C) character, the comma must be escaped with a double reverse solidus character (\; U+005C). This double escape is required, because both the routine splitting 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" would 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 in such an option. For example the option time-servers allows the specification of more than one IPv4 address, so allowing clients to obtain the addresses of multiple NTP servers. The describes the configuration syntax to create custom option definitions (formats). It is generally not allowed to create custom definitions for standard options, even if the definition being created matches the actual option format defined in the RFCs. There is an exception from this rule for standard options for which Kea currently does not provide a definition. In order to use such options, a server administrator must create a definition as described in in the 'dhcp4' option space. This definition should match the option format described in the relevant RFC but the configuration mechanism will allow any option format as it presently has no means to validate it. Name Code Type Array? Returned if not requested? time-offset2int32falsefalserouters3ipv4-addresstruetruetime-servers4ipv4-addresstruefalsename-servers5ipv4-addresstruefalsedomain-name-servers6ipv4-addresstruetruelog-servers7ipv4-addresstruefalsecookie-servers8ipv4-addresstruefalselpr-servers9ipv4-addresstruefalseimpress-servers10ipv4-addresstruefalseresource-location-servers11ipv4-addresstruefalseboot-size13uint16falsefalsemerit-dump14stringfalsefalsedomain-name15fqdnfalsetrueswap-server16ipv4-addressfalsefalseroot-path17stringfalsefalseextensions-path18stringfalsefalseip-forwarding19booleanfalsefalsenon-local-source-routing20booleanfalsefalsepolicy-filter21ipv4-addresstruefalsemax-dgram-reassembly22uint16falsefalsedefault-ip-ttl23uint8falsefalsepath-mtu-aging-timeout24uint32falsefalsepath-mtu-plateau-table25uint16truefalseinterface-mtu26uint16falsefalseall-subnets-local27booleanfalsefalsebroadcast-address28ipv4-addressfalsefalseperform-mask-discovery29booleanfalsefalsemask-supplier30booleanfalsefalserouter-discovery31booleanfalsefalserouter-solicitation-address32ipv4-addressfalsefalsestatic-routes33ipv4-addresstruefalsetrailer-encapsulation34booleanfalsefalsearp-cache-timeout35uint32falsefalseieee802-3-encapsulation36booleanfalsefalsedefault-tcp-ttl37uint8falsefalsetcp-keepalive-interval38uint32falsefalsetcp-keepalive-garbage39booleanfalsefalsenis-domain40stringfalsefalsenis-servers41ipv4-addresstruefalsentp-servers42ipv4-addresstruefalsevendor-encapsulated-options43emptyfalsefalsenetbios-name-servers44ipv4-addresstruefalsenetbios-dd-server45ipv4-addresstruefalsenetbios-node-type46uint8falsefalsenetbios-scope47stringfalsefalsefont-servers48ipv4-addresstruefalsex-display-manager49ipv4-addresstruefalsedhcp-option-overload52uint8falsefalsedhcp-server-identifier54ipv4-addressfalsetruedhcp-message56stringfalsefalsedhcp-max-message-size57uint16falsefalsevendor-class-identifier60hexfalsefalsenwip-domain-name62stringfalsefalsenwip-suboptions63hexfalsefalsenisplus-domain-name64stringfalsefalsenisplus-servers65ipv4-addresstruefalsetftp-server-name66stringfalsefalseboot-file-name67stringfalsefalsemobile-ip-home-agent68ipv4-addresstruefalsesmtp-server69ipv4-addresstruefalsepop-server70ipv4-addresstruefalsenntp-server71ipv4-addresstruefalsewww-server72ipv4-addresstruefalsefinger-server73ipv4-addresstruefalseirc-server74ipv4-addresstruefalsestreettalk-server75ipv4-addresstruefalsestreettalk-directory-assistance-server76ipv4-addresstruefalseuser-class77hexfalsefalseslp-directory-agent78record (boolean, ipv4-address)truefalseslp-service-scope79record (boolean, string)falsefalsends-server85ipv4-addresstruefalsends-tree-name86stringfalsefalsends-context87stringfalsefalsebcms-controller-names88fqdntruefalsebcms-controller-address89ipv4-addresstruefalseclient-system93uint16truefalseclient-ndi94record (uint8, uint8, uint8)falsefalseuuid-guid97record (uint8, hex)falsefalseuap-servers98stringfalsefalsegeoconf-civic99hexfalsefalsepcode100stringfalsefalsetcode101stringfalsefalsenetinfo-server-address112ipv4-addresstruefalsenetinfo-server-tag113stringfalsefalsedefault-url114stringfalsefalseauto-config116uint8falsefalsename-service-search117uint16truefalsesubnet-selection118ipv4-addressfalsefalsedomain-search119fqdntruefalsevivco-suboptions124hexfalsefalsevivso-suboptions125hexfalsefalsepana-agent136ipv4-addresstruefalsev4-lost137fqdnfalsefalsecapwap-ac-v4138ipv4-addresstruefalsesip-ua-cs-domains142fqdntruefalserdnss-selection146record (uint8, ipv4-address, ipv4-address, fqdn)truefalsev4-portparams159record (uint8, uint8, uint16)falsefalsev4-captive-portal160stringfalsefalseoption-6rd212record (uint8, uint8, ipv6-address, ipv4-address)truefalsev4-access-domain213fqdnfalsefalse NameMeaninghexAn arbitrary string of bytes, specified as a set of hexadecimal digits.booleanBoolean value with allowed values true or falseemptyNo value, data is carried in suboptionsfqdnFully qualified domain name (e.g. www.example.com)ipv4-addressIPv4 address in the usual dotted-decimal notation (e.g. 192.0.2.1)ipv6-addressIPv6 address in the usual colon notation (e.g. 2001:db8::1)ipv6-prefixIPv6 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 valuepsidPSID 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 the details about the PSID wire representationrecordStructured data that may be comprised of any types (except "record" and "empty"). The array flag applies to the last field only.stringAny texttupleA length encoded as a 8 (16 for DHCPv6) bit unsigned integer followed by a string of this lengthuint88 bit unsigned integer with allowed values 0 to 255uint1616 bit unsigned integer with allowed values 0 to 65535uint3232 bit unsigned integer with allowed values 0 to 4294967295int88 bit signed integer with allowed values -128 to 127int1616 bit signed integer with allowed values -32768 to 32767int3232 bit signed integer with allowed values -2147483648 to 2147483647

Kea supports custom (non-standard) DHCPv4 options. Assume that we want to define a new DHCPv4 option called "foo" which will have a code 222 and will convey a single unsigned 32 bit integer value. We can define such an option by using 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. This should be non-empty if the 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 it should be left blank. Note that the above set of comments define the format of the new option and do not set its values. The name, code and type parameters are required, all others are optional. The array default value is false. The record-types and encapsulate default values are blank (i.e. ""). 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 simple use of primitives (uint8, string, ipv4-address etc): it is possible to define an option comprising a number of existing primitives. Assume 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 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 . The values of the option are set as follows: "Dhcp4": { "option-data": [ { "name": "bar", "space": "dhcp4", "code": 223, "csv-format": true, "data": "192.0.2.100, 123, true, Hello World" } ], ... } csv-format is set to true to indicate that the data field comprises a command-separated list of values. The values in the 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. In the general case, boolean values are specified as true or false, without quotes. Some specific boolean parameters may accept also "true", "false", 0, 1, "0" and "1". Future versions of Kea will accept all those values for all boolean parameters. 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).

Options with code between 224 and 254 are reserved for private use. They can be defined at the global scope or at client class local scope: this allows to use option definitions depending on context and to set option data accordingly. For instance to configure an old PXEClient vendor: "Dhcp4": { "client-class": [ { "name": "pxeclient", "test": "option[vendor-class-identifier].text == 'PXEClient'", "option-def": [ { "name": "configfile", "code": 209, "type": "string" } ], ... }, ... ], ... } As the Vendor Specific Information option (code 43) has vendor specific format, i.e. can carry either raw binary value or sub-options, this mechanism is available for this option too. In the following example taken from a real configuration two vendor classes use the 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-class": [ { "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: The local definition of a client class the incoming packet belongs to If none, the global definition If none, the last resort definition described in the next section (backward compatible with previous Kea versions). This last resort definition for the Vendor Specific Information option (code 43) is not compatible with a raw binary value. So when there are some known cases where a raw binary value will be used, a client class must be defined with a classification expression matching these cases and an option definition for the VSI option with a binary type and no encapsulation. Option definitions in client classes is allowed only for these limited option set (codes 43 and from 224 to 254), and only for DHCPv4.

Currently there are two option spaces defined for the DHCPv4 daemon: "dhcp4" (for the top level DHCPv4 options) and "vendor-encapsulated-options-space", which is empty by default but in which options can be defined. Such options will be carried in the Vendor Specific Information option (code 43). The following examples show how to define an option "foo" in that space that has a code 1, and 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" } ], ... } We also include the Vendor Specific Information option, the option that conveys our sub-option "foo". This is required, else 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 possibility, added in Kea 1.3, is to redefine the option, see .

It is sometimes useful to define a completely new option space. This is the case when user creates 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 will have a separate numbering scheme and may overlap with the codes of standard options. Note that creation of a new option space when defining sub-options for a standard option is not required, because it is created by default if the standard option is meant to convey any sub-options (see ). Assume that we want to have a DHCPv4 option called "container" with code 222 that conveys two sub-options with codes 1 and 2. First we 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 our 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" } ], ... } Note that 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 was required to carry an 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.

In many cases it is not required to specify all parameters for an option configuration and the default values may 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 an option name or option code to identify an option. If this parameter is unspecified, the option code must be specified. code - the server requires an option name or 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 has its definition (it is either a standard option or an administrator created a 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 option code. space - if the option space is unspecified it will default to 'dhcp4' which is an option space holding DHCPv4 standard 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 which the specification allows for the 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 will assume that the option data is specified as a list of comma separated values to be assigned to individual fields of the DHCP option. This behavior has changed in Kea 1.2. Older versions used additional logic to determine whether the csv-format should be true or false. That is no longer the case.

The DHCPv4 server supports the 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, e.g. addresses of DNS servers. In order 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 will respond to the DHCPINFORM when the client is associated with a subnet defined in the server's configuration. An example subnet configuration will look like this: "Dhcp4": { "subnet4": [ { "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. Note that the subnet definition does not require the address pool configuration if it will be used solely for the 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, 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.

The DHCPv4 server includes support for client classification. For a deeper discussion of the classification process see . In certain cases it is useful to differentiate between different types of clients and treat them accordingly. It is envisaged that client classification will be used for changing the behavior of almost any part of the DHCP message processing, including the assignment of leases from different pools, the assignment of different options (or different values of the same options) etc. In the current release of the software however, there are only three mechanisms that take advantage of client classification: subnet selection, assignment of different options, and, for cable modems, there are 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. Here, 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 that should get a lease from subnet B. That segregation is essential to prevent overly curious users from playing with their cable modems. For details on how to set up class restrictions on subnets, see . The process of doing classification is conducted in three 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. The third step is to assign options, again possibly based on the class information. There are two methods of doing classification. The first is automatic and relies on examining the values in the vendor class options. 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 allows you to specify an expression that is evaluated for each packet. If the result is true the packet is a member of the class. Care should be taken with client classification as it is easy for clients that do not meet class criteria to be denied any service altogether.

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 (that conveys an IPv4 address, which is set in the siaddr field), server-hostname (that conveys a server hostname, can be up to 64 bytes long and will be sent in the sname field) and boot-file-name (that conveys the configuration file, can be up to 128 bytes long and will be sent using file field). Obviously, there are many ways to assign clients to specific classes, but for the PXE clients the client architecture type option (code 93) seems to be particularly suited to make the distinction. The following example checks if the client identifies itself as PXE device with architecture EFI x86-64, and sets several fields if it does. See Section 2.1 of RFC 4578) or the documentation of your client 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 there are multiple classes defined and an incoming packet is matched to multiple classes, the class whose name is alphabetically the first is used.

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_", it is interpreted as a class. For example, modern cable modems will send this option with value "docsis3.0" and as a result the packet will belong to class "VENDOR_CLASS_docsis3.0". Kea 1.0 and earlier versions performed special actions for clients that were in VENDOR_CLASS_docsis3.0. This is no longer the case in Kea 1.1 and later. In these versions the old behavior 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 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 docsis3.0 client class are allowed to use that pool. "Dhcp4": { "subnet4": [ { "subnet": "192.0.2.0/24", "pools": [ { "pool": "192.0.2.10 - 192.0.2.20" } ], "client-class": "VENDOR_CLASS_docsis3.0" } ], ... }

The following example shows how to configure a class using an expression and a subnet that makes use of the class. This configuration defines the class named "Client_foo". It is comprised of all clients who's 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": [ { "subnet": "192.0.2.0/24", "pools": [ { "pool": "192.0.2.10 - 192.0.2.20" } ], "client-class": "Client_foo" }, ... ], ... }

As mentioned earlier, kea-dhcp4 can be configured to generate requests to the DHCP-DDNS server (referred to here as "D2" ) to update DNS entries. These requests are known as NameChangeRequests or NCRs. Each NCR contains the following information: Whether it is a request to add (update) or remove DNS entries Whether the change requests forward DNS updates (A records), reverse DNS updates (PTR records), or both. The FQDN, lease address, and DHCID The parameters for controlling the generation of NCRs for submission to D2 are contained in the dhcp-ddns section of the kea-dhcp4 server configuration. The mandatory parameters for the DHCP DDNS configuration are enable-updates which is unconditionally required, and qualifying-suffix which has no default value and is required when enable-updates is set to true. The two (disabled and enabled) minimal DHCP DDNS configurations are: "Dhcp4": { "dhcp-ddns": { "enable-updates": false }, ... } and for example: "Dhcp4": { "dhcp-ddns": { "enable-updates": true, "qualifying-suffix": "example." }, ... } The default values for the "dhcp-ddns" section are as follows: "server-ip": "127.0.0.1" "server-port": 53001 "sender-ip": "" "sender-port": 0 "max-queue-size": 1024 "ncr-protocol": "UDP" "ncr-format": "JSON" "override-no-update": false "override-client-update": false "replace-client-name": "never" "generated-prefix": "myhost"

In order 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 - determines whether or not kea-dhcp4 will generate NCRs. By default, this value is false hence DDNS updates are disabled. To enable DDNS updates set this value to true: server-ip - IP address on which D2 listens for requests. The default is the local loopback interface at address 127.0.0.1. You may specify either an IPv4 or IPv6 address. server-port - port on which D2 listens for requests. The default value is 53001. sender-ip - IP address which kea-dhcp4 should use to send requests to D2. The default value is blank which instructs kea-dhcp4 to select a suitable address. sender-port - port which kea-dhcp4 should use to send requests to D2. The default value of 0 instructs kea-dhcp4 to select a suitable port. max-queue-size - maximum number of requests allowed to queue 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 will be turned off until the queue backlog has been sufficiently reduced. The intention 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 - socket protocol use when sending requests to D2. Currently only UDP is supported. TCP may be available in an upcoming release. ncr-format - packet format to use when sending requests to D2. Currently only JSON format is supported. Other formats may be available in future releases. 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 would be required: "Dhcp4": { "dhcp-ddns": { "server-ip": "192.168.1.10", "server-port": 900, ... }, ... }

kea-dhcp4 follows the behavior prescribed for DHCP servers in RFC 4702. It is important to keep in mind that kea-dhcp4 provides the initial decision making 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 (). This section describes when kea-dhcp4 will generate NCRs and the configuration parameters that can be used to influence this decision. It assumes that the enable-updates parameter is true. In general, kea-dhcp4 will generate DDNS update requests when: A new lease is granted in response to a DHCP REQUEST An existing lease is renewed but the FQDN associated with it has changed. An existing lease is released in response to a DHCP RELEASE In the second case, lease renewal, two DDNS requests will be issued: one request to remove entries for the previous FQDN and a second request to add entries for the new FQDN. In the last case, a lease release, a single DDNS request to remove its entries will be made. The decision making involved when granting a new lease (the first case) is more involved. When a new lease is granted, kea-dhcp4 will generate a DDNS update request if the DHCP REQUEST contains either the FQDN option (code 81) or the Host Name option (code 12). If both are present, the server will use the FQDN option. By default kea-dhcp4 will respect the FQDN N and S flags specified by the client as shown in the following table: 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 will honor the client's wishes and generate a DDNS request to the D2 server to update only reverse DNS data. The parameter override-client-update can be used to instruct the server to override client delegation requests. When this parameter is true, kea-dhcp4 will disregard requests for client delegation and generate 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": { "dhcp-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, override-no-update, can be used to instruct the server to disregard the client's wishes. When this parameter is true, kea-dhcp4 will generate 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, the following values should be set in your configuration: "Dhcp4": { "dhcp-ddns": { "override-no-update": true, ... }, ... } kea-dhcp4 will always generate DDNS update requests if the client request only contains the Host Name option. In addition it will include 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 will be the name submitted to D2 in the DDNS update request.

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 upon what it receives from the client in the DHCP REQUEST. The default rules for constructing the FQDN that will be used for DNS entries are: If the DHCPREQUEST contains the client FQDN option, the candidate name is taken from there, otherwise it is taken from the Host Name option. 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. If the candidate name provided is empty, generate a FQDN using a configurable prefix and suffix. If the client provided neither option, then no DNS action will be taken. These rules can amended by setting the 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 that formerly, this parameter was a boolean and permitted only values of true and false. Boolean values have been deprecated and are no longer accepted. If you are currently using booleans, you 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 replace-client-name to always as follows: "Dhcp4": { "dhcp-ddns": { "replace-client-name": "always", ... }, ... } The prefix used in the generation of a FQDN is specified by the generated-prefix parameter. The default value is "myhost". To alter its value, simply set it to the desired string: "Dhcp4": { "dhcp-ddns": { "generated-prefix": "another.host", ... }, ... } The suffix used when generating a FQDN or when qualifying a partial name is specified by the qualifying-suffix parameter. This parameter has no default value, thus it is mandatory when DDNS updates are enabled. To set its value simply set it to the desired string: "Dhcp4": { "dhcp-ddns": { "qualifying-suffix": "foo.example.org", ... }, ... } When generating a name, kea-dhcp4 will construct name of the format: [generated-prefix]-[address-text].[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 "example.com", and the default value is used for generated-prefix, the generated FQDN would be: myhost-172-16-1-10.example.com.

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 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 subnet can be set to 0.0.0.0, which means that next-server should not be sent. It may also be set to an empty string, which means the same as if it was not defined at all, i.e. use the global value. The server-hostname (that conveys a server hostname, can be up to 64 bytes long and will be sent in the sname field) and boot-file-name (that conveys the configuration file, can be up to 128 bytes long and will be sent using 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", ... } ] }

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 confused clients that did not have MAC address or client-id; see RFC 6842. for details. That behavior has changed with the publication of RFC 6842 which updated RFC 2131. That update states that the server must send client-id if the client sent it. 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, ... }

The DHCP server must be able to identify the client (and distinguish it from other clients) from which it receives the message. There are many reasons why this identification is required and 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 is trying 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 the dual stack networks there is often a need to correlate the lease information stored in DHCPv4 and DHCPv6 server for a particular host. Using common identification information by the DHCPv4 and DHCPv6 client allows the network administrator to achieve this correlation and better administer the network. DHCPv4 makes use of 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 "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 is the one which has obtained the lease. Moreover, RFC 4361 gives the recommendation to use a DUID (see RFC 3315, the DHCPv6 specification) carried as "client identifier" when dual stack networks are in use to provide consistent identification information of the client, regardless of the protocol type it is using. Kea adheres to these specifications and the "client identifier" by default takes precedence over the value carried in "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 will try to find out if the client already has a lease in the database and will hand out that lease rather than allocate a new one. Each lease in the lease database is associated with the "client identifier" and/or "chaddr". The server will first use the "client identifier" (if present) to search the lease. If the lease is found, the server will treat 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 will perform another lookup using the "chaddr". If this lookup returns no result, the client is considered as not having a lease and the new lease will be 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 "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 by instructing the server to disregard 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": [ { "subnet": "192.0.10.0/24", "pools": [ { "pool": "192.0.2.23-192.0.2.87" } ], "match-client-id": false }, { "subnet": "10.0.0.0/8", "pools": [ { "pool": "10.0.0.23-10.0.2.99" } ], } ] } The match-client-id 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. The false means that the server will only use the "chaddr" to search for client's lease. Whether the DHCID for DNS updates is generated from the "client identifier" or "chaddr" is controlled through the same parameter accordingly. 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 explain what happens when the client obtains its lease for one setting of the match-client-id and then renews when the setting has been changed. First consider the case when the client obtains the lease when the match-client-id is set to true. The server will store the lease information including "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 hasn't changed its MAC address the server should successfully find the existing lease. The "client identifier" associated with the returned lease is ignored and the client is allowed to use this lease. When the lease is renewed only the "chaddr" is 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 setting is changed to match-client-id set 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 "client identifier" which is different from the "client identifier" used by the client the lease will be assumed to belong to another client and the new lease will be allocated.

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 support is experimental and the details of the inter-process communication can 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. In order to address this issue three new configuration entries have been added. 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. The following configuration was used during some tests: { # DHCPv4 conf "Dhcp4": { "interfaces-config": { "interfaces": [ "eno33554984" ] }, "lease-database": { "type": "memfile", "name": "leases4" }, "valid-lifetime": 4000, "subnet4": [ { "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 }, "Logging": { "loggers": [ { "name": "kea-dhcp4", "output_options": [ { "output": "/tmp/kea-dhcp4.log" } ], "severity": "DEBUG", "debuglevel": 0 } ] } }

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 will receive the same address from the server every time, and other clients will generally not receive that address. Another example when the host reservations are applicable is 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. Note that there may be cases when the new reservation has been made for the client for the address being currently in use by another client. We call this situation a "conflict". The conflicts get resolved automatically over time as described in subsequent sections. Once the conflict is resolved, the client will keep receiving the reserved configuration when it renews. Host reservations are defined as parameters for each subnet. Each host has to be identified by an identifier, for example the hardware/MAC address. There is an optional reservations array in the Subnet4 element. Each element in that array is a structure that holds information about reservations for a single host. In particular, the structure has to have 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 DHCPv4 message such as siaddr, sname or file. In Kea 1.0.0 it was only possible to create host reservations using client's hardware address. Host reservations by client identifier, DUID and circuit-id have been added in Kea 1.1.0. The following example shows how to reserve addresses for specific hosts: "subnet4": [ { "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. (Note that if you plan to do DNS updates, it is strongly recommended for the hostnames to be unique.) The third example reserves address 192.0.3.203 to a client whose request would be relayed by a relay agent that inserts a circuit-it 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 and in actual deployments it is recommended to use as few types as possible (preferably just one). See 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 it is expected to visit. It is not allowed 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 reservation 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 also supports host reservation has to perform additional checks: not only if 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. there is a reservation for it). That additional check incurs additional overhead.

In a typical scenario there is an IPv4 subnet defined, e.g. 192.0.2.0/24, with 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. However, upcoming releases may offer improved performance if there are only out-of-pool reservations as the server will be able to skip reservation checks when dealing with existing leases. Therefore, system administrators are encouraged to use out-of-pool reservations if possible.

As the 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. Let's analyze the situation here. If Host B boots up and requests an address, the server is not able to assign the reserved address 192.0.2.10. A naive approach would to be immediately remove the existing lease for the Host A and create a new one for the Host B. That would not solve the problem, though, because as soon as the Host B gets the address, it will detect that the address is already in use by the Host A and would send the DHCPDECLINE message. Therefore, in this situation, the server has to temporarily assign a different address (not matching what has been reserved) to the 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. Therefore the server will inform the 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 small chance that the DHCPNAK may be lost 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 the 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 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 out that there is a reservation for that host and 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 will attempt to get the reserved address. Had the Host C requested address 192.0.2.10 after the reservation was made, the server will either offer a different address (when responding to DHCPDISCOVER) or would send DHCPNAK (when responding to DHCPREQUEST). This recovery mechanism allows the server to fully recover from a case where reservations conflict with the existing leases. This procedure takes time and will roughly take as long as the value set for of renew-timer. The best way to avoid such recovery is to not 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 could get this address.

When the reservation for a client includes the hostname, the server will return this hostname to the client in the Client FQDN or Hostname options. The server responds with the Client FQDN option only if the client has included Client FQDN option in its message to the server. The server will respond with the Hostname option if the client included Hostname option in its message to the server or when the client requested Hostname option using Parameter Request List option. The server will return 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 qualifying-suffix parameter. For example, the following subnet configuration: { "subnet4": [ { "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" } ] }], "dhcp-ddns": { "enable-updates": true, "qualifying-suffix": "example.isc.org." } } will result in assigning the "alice-laptop.example.isc.org." hostname to the client using the MAC address "aa:bb:cc:dd:ee:ff". If the 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 fully qualified name. Thus, by leaving the qualifying-suffix empty it is possible to qualify hostnames for the different clients with different domain names: { "subnet4": [ { "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, } }

Kea 1.1.0 introduced the ability to specify options on a per host basis. The options follow the same rules as any other options. These can be standard options (see ), custom options (see ) or vendor specific options (see ). 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 on host level have the highest priority. In other words, if there are options defined with the same type on global, subnet, class and host level, the host specific values will be used.

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, e.g. reserved IPv4 address. These parameters are optional, i.e. a subset of them can specified, or all of them can be omitted.

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 reservations mechanisms also allow for statically assigning classes to the clients. The definitions of these classes must exist in the Kea configuration. The following configuration snippet shows how to specify that a client belongs to classes reserved-class1 and reserved-class2. Those classes are associated with specific options being 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": [ { "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" ] } ] } ] } Static class assignments, as shown above, can be used in conjunction with classification using expressions.

It is possible to store host reservations in MySQL, PostgreSQL or Cassandra. See for information on how to configure Kea to use reservations stored in MySQL, PostgreSQL or Cassandra. Kea provides dedicated hook for managing reservations in a database, section provide detailed information. In Kea maximum length of an option specified per host is arbitrarily set to 4096 bytes.

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 not be used by another DHCP client. It also must not be reserved for another client. Second, when renewing a lease, additional check must be performed whether the address being renewed is not reserved for another client. Finally, when a host renews an address, the server has to check whether there is a reservation for this host, so 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 the 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. Allowed values are: all - enables all host reservation types. This is the default value. This setting is the safest and the most flexible. It allows in-pool and out-of-pool reservations. As all checks are conducted, it is also the slowest. out-of-pool - allows only out of pool host reservations. With this setting in place, the server may assume 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 of your reservations use in-pool address. Caution is advised when using this setting: Kea does not sanity check the reservations against reservation-mode and misconfiguration may cause problems. disabled - host reservation support is disabled. As there are no reservations, the server will skip all checks. Any reservations defined will be completely ignored. As the checks are skipped, the server may operate faster in this mode. An example configuration that disables reservation looks like follows: "Dhcp4": { "subnet4": [ { "subnet": "192.0.2.0/24", "reservation-mode": "disabled", ... } ] } Another aspect of the host reservations are the different types of identifiers. Kea 1.1.0 supports four types of identifiers (hw-address, duid, client-id and circuit-id), but more identifier types are likely to be added in the future. This is beneficial from a usability perspective. However, there is a drawback. For each incoming packet Kea has to to extract each identifier type and then query the database to see if there is a reservation done 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 has been introduced. It takes a list of identifier types as a parameter. Kea will check 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 limited to one. If your 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 number of queries. An example of host reservation identifiers 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" ]

DHCP servers use subnet information in two ways. First, it is used to determine the point of attachment, or simply put, where the client is connected to the network. Second, the subnet information is used to group information pertaining to specific location in the network. This approach works well in general case, but the are scenarios where the boundaries are blurred. Sometimes it is useful to have more than one logical IP subnet being deployed on the same physical link. The need to understand that two or more subnets are used on the same link requires additional logic in the DHCP server. This capability has been added in Kea 1.3.0. It is called "shared networks" in Kea and ISC DHCP projects. It is sometimes also called "shared subnets". In Microsoft's nomenclature it is called "multinet". There are many use cases where the feature is useful. This paragraph explains just a handful of the most common ones. The first and by far the most common use case is an existing 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 subnet on top of the existing one. Sometimes, due to address space fragmentation (e.g. only many disjoint /24s are available) this is the only choice. Also, configuring additional subnet 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 subnet for cable modems to prevent users from tinkering with their cable modems. In this case, the distinction is based on the type of device, rather than address space exhaustion. A client connected to a shared network may be assigned an address from any of the address 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 pools exhaustion) it will use another subnet from the same shared network and try to allocate an address from this subnet etc. Therefore, in the typical case, the server will allocate all addresses available for 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 the other subnets before the address pools in the first subnet get exhausted, e.g. when the client provides a hint that belongs to another subnet or the client has reservations in a different than default subnet. It is strongly discouraged for the Kea deployments to assume that the server doesn't allocate addresses from other subnets until it uses all the addresses from the first subnet in the shared network. Apart from the fact that hints, host reservations and client classification affect subnet selection, it is also foreseen that we will enhance allocation strategies for shared networks in the future versions of Kea, so as the selection of subnets within a shared network is equally probable (unpredictable). In order 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", // Subnet selector can be specifed on the shared network level. // Subnets from this shared network will be selected for directly // connected clients sending requests to server's "eth0" interface. "interface": "eth0", // This starts a list of subnets in this shared network. // There are two subnets in this example. "subnet4": [ { "subnet": "10.0.0.0/8", "pools": [ { "pool": "10.0.0.1 - 10.0.0.99" } ], }, { "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 your network you'll have a mix of regular, // "plain" subnets and shared networks. It is perfectly valid to mix // them in the same config file. // // This is regular subnet. It's not part of any shared-network. "subnet4": [ { "subnet": "192.0.3.0/24", "pools": [ { "pool": "192.0.3.1 - 192.0.3.200" } ], "interface": "eth1" } ] } // end of Dhcp4 } As you see 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 ID for subnets, but gives you more flexibility. This 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 it is allowed to define a shared network with just one subnet or even an empty one. This is not a recommended practice in production networks, as the shared network logic requires additional processing and thus lowers server's performance. To avoid unnecessary performance degradation the shared subnets should only be defined when required by the deployment. Shared networks provide an ability to specify many parameters in the shared network scope that will apply to all subnets within it. If necessary, you can specify a parameter on 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": [ { "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" } ] }, { "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, a valid lifetime is set to 10 minutes (600s). However, the first subnet overrides some of the values (valid lifetime is 20 minutes, different IP address for log-servers), but also adds its own option (router address). Assuming a client asking for router and log servers options is assigned a lease from this subnet, he will get a lease for 20 minutes and log-servers and routers value of 10.0.0.254. If the same client is assigned to the second subnet, he will get a 10 minutes long lease, log-servers value of 1.2.3.4 and routers set to 192.0.2.1.

It is possible to specify interface name in the shared network scope to tell the server that this specific shared network is reachable directly (not via relays) using local network interface. It is sufficient to specify it once on the shared network level. 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. It is allowed to specify interface parameter on each subnet, although its value must be the same for each subnet. Thus it's usually more convenient to specify it once on the shared network level. "shared-networks": [ { "name": "office-floor-2", // This tells Kea that the whole shared networks is reachable over // local interface. This applies to all subnets in this network. "interface": "eth0", "subnet4": [ { "subnet": "10.0.0.0/8", "pools": [ { "pool": "10.0.0.1 - 10.0.0.99" } ], "interface": "eth0" }, { "subnet": "192.0.2.0/24", "pools": [ { "pool": "192.0.2.100 - 192.0.2.199" } ] // Specifying a different interface name is configuration // error: // "interface": "eth1" } ] } ] Somewhat similar to interface names, also relay IP addresses can be specified for the whole shared network. However, depending on your relay configuration, it may use different IP addresses depending on which subnet is being used. Thus there is no requirement to use the same IP relay address for each subnet. Here's an example: "shared-networks": [ {" "name": "kakapo", "relay": { "ip-address": "192.3.5.6" }, "subnet4": [ { "subnet": "192.0.2.0/26", "relay": { "ip-address": "192.1.1.1" }, "pools": [ { "pool": "192.0.2.63 - 192.0.2.63" } ] }, { "subnet": "10.0.0.0/24", "relay": { "ip-address": "192.2.2.2" }, "pools": [ { "pool": "10.0.0.16 - 10.0.0.16" } ] } ] } ] In this particular case the relay IP address specified on network level doesn't have much sense, as it is overridden in both subnets, but it was left there as an example of how one could be defined on network level. Note that the relay agent IP address typically belongs to the subnet it relays packets from, but this is not a strict requirement. Therefore Kea accepts any value here as long as it is valid IPv4 address.

Sometimes it is desired to segregate clients into specific subnets based on some properties. This mechanism is called client classification and is described in . 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 the clients when client classification is in use, to assure that the desired subnet is selected for a given client type. If a subnet is associated with some classes, only the clients belonging to any of these classes 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 the subnet including client classes 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": [ { "subnet": "192.0.2.0/26", "pools": [ { "pool": "192.0.2.1 - 192.0.2.63" } ], }, { "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 "b-devices" class (because it includes option 93 with a value of 0x0002) it doesn't guarantee that the subnet 10.0.0.0/24 will be used (or preferred) for this client. The server can use any 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 pereceived as a way to restrict access to certain subnets, rather than a way to express subnet preference. For example, if the client doesn't 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 the 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 required to apply classification on all subnets. The following example defines two classes of devices. 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": [ { "subnet": "192.0.2.0/26", "pools": [ { "pool": "192.0.2.1 - 192.0.2.63" } ], "client-class": "a-devices" }, { "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" will be able to use subnet 192.0.2.0/26 and only clients belonging to b-devices will be able to use subnet 10.0.0.0/24. Care should be taken to not define too restrictive classification rules, as clients that are unable to use any subnets will be refused service. Although, this may be a desired outcome if one desires to service only clients of known properties (e.g. only VoIP phones allowed on a given link). Note that it is possible to achieve similar effect as 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. "interface" or "relay" parameter. All subnets belonging to this shared network will inherit those parameters.

Subnets being 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 if there's a reservation for a given client in his initially selected subnet, it goes through all subnets in a shared network looking 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 will be picked for the client. Although it's technically not an error, it is considered a 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 you plan to use database storage for host reservations. If ID is not specified, the values for it be autogenerated, i.e. it will assign increasing integer values starting from 1. Thus, the autogenerated IDs are not stable across configuration changes.

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 will use multiple server identifiers if it is receiving queries on multiple interfaces. It is possible to override default server identifier values by specifying "dhcp-server-identifier" option. This option is only supported on the global, shared network and subnet level. It must not be specified on client class and host reservation level. The following example demonstrates how to override server identifier for a subnet: "subnet4": [ { "subnet": "192.0.2.0/24", "option-data": [ { "name": "dhcp-server-identifier", "data": "10.2.5.76" } ], ... } ]

The DHCPv4 server differentiates between the directly connected clients, clients trying to renew leases and clients sending their messages through relays. For directly connected clients, the server will check the configuration for the interface on which the message has been received and, if the server configuration doesn't match any configured subnet, the message is discarded. Assuming that the server's interface is configured with the IPv4 address 192.0.2.3, the server will only process messages received through this interface from a directly connected client if there is a subnet configured to which this IPv4 address belongs, e.g. 192.0.2.0/24. The server will use this subnet to assign 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. 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 for details. 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.

A relay has to 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 that belongs to the subnet from which the server will assign addresses. In the typical case, 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 where this is the case are long lasting network renumbering (where both old and new address space is 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 will be able to select this subnet for any incoming packets that came from a relay that has an address in 192.0.2.0/24 subnet. It will also select that subnet for a relay with address 10.0.0.1. "Dhcp4": { "subnet4": [ { "subnet": "192.0.2.0/24", "pools": [ { "pool": "192.0.2.10 - 192.0.2.20" } ], "relay": { "ip-address": "10.0.0.1" }, ... } ], ... } If "relay" is specified, the "ip-address" parameter within it is mandatory.

In certain cases, it is useful to mix relay address information, introduced in with client classification, explained in . One specific example is cable network, where typically modems get addresses from a different subnet than all devices connected behind them. Let us assume that there is one CMTS (Cable Modem Termination System) 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 modems should get addresses from another subnet (192.0.2.0/24). The CMTS that acts as a relay uses address 10.1.1.1. The following configuration can serve that configuration: "Dhcp4": { "subnet4": [ { "subnet": "10.1.1.0/24", "pools": [ { "pool": "10.1.1.2 - 10.1.1.20" } ], "client-class" "docsis3.0", "relay": { "ip-address": "10.1.1.1" } }, { "subnet": "192.0.2.0/24", "pools": [ { "pool": "192.0.2.10 - 192.0.2.20" } ], "relay": { "ip-address": "10.1.1.1" } } ], ... }

The DHCPv4 server is configured with a certain pool of addresses that it is expected to hand out to the DHCPv4 clients. It is assumed that the server is authoritative and has complete jurisdiction over those addresses. However, due to 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 will do a sanity check (if the client declining an address really was supposed to use it), and then will conduct a clean up operation. Any DNS entries related to that address will be removed, the fact will be logged and hooks will be triggered. After that is done, the address will be marked as declined (which indicates that it is used by an unknown entity and thus not available for assignment to anyone) and a probation time will be set on it. Unless otherwise configured, the probation period lasts 24 hours. After that period, 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. On the other hand, it provides an opportunity to recover from such an event automatically, without any sysadmin 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 will instruct the server to recycle declined leases after an hour. There are several statistics and hook points associated with the Decline handling procedure. The lease4_decline hook 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 global and subnet specific variants). (See and 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. Note about statistics: The 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 simply do assigned-addresses/total-addresses. This would have a bias towards under-representing pool utilization. As this has a potential for major issues, we decided not to decrease assigned addresses immediately after receiving DHCPDECLINE, but to do it later when we recover the address back to the available pool.

This section describes DHCPv4-specific statistics. For a general overview and usage of statistics, see . The DHCPv4 server supports the following 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 your 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. Non-zero value indicates an error. One likely cause would be a misbehaving relay agent that incorrectly forwards DHCPOFFER messages towards the server, rather 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 server's response (DHCPOFFER), accepted it and 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. Non-zero value indicates an error. One likely cause would be a misbehaving relay agent that incorrectly forwards DHCPACK messages towards the server, rather 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. Non-zero value indicates an error. One likely cause would be a misbehaving relay agent that incorrectly forwards DHCPNAK messages towards the server, rather 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 stop using 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 that leased an address, but discovered that the address is currently used by an unknown device in your 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. Non-zero value of this statistic indicates that the server received a packet that it wasn't able to recognize: either with unsupported type or possibly malformed (without 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 server to respond. There are exceptions (e.g. DHCPRELEASE), so do not worry, if it is lesser 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 is 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 to not 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 malformed or truncated packet. This may indicate problems in your 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, 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 The total number of addresses available for DHCPv4 management. In other words, this is the sum of all addresses in all configured pools. This statistic changes only during configuration changes. Note 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. This statistic is exposed for each subnet separately. This statistic is reset during reconfiguration event. subnet[id].assigned-addresses integer This statistic shows the 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 is decreased 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. This statistic is reset during reconfiguration event. reclaimed-leases integer This statistic is the number of expired leases that have been reclaimed since server startup. It is incremented each time an expired lease is reclaimed and is reset when the server is reconfigured. subnet[id].reclaimed-leases integer This statistic is the number of expired leases associated with a given subnet (id is the subnet-id) that have been reclaimed since server startup. It is incremented each time an expired lease is reclaimed and is reset when the server is reconfigured. declined-addresses integer This statistic shows the number of IPv4 addresses that are currently declined, so counting the number of leases currently unavailable. Once a lease is recovered, this statistic will be decreased. Ideally, this statistic should be zero. If this statistic is non-zero (or worse increasing), a network administrator should investigate if there is a misbehaving device in his network. This is a global statistic that covers all subnets. subnet[id].declined-addresses integer This statistic shows the number of IPv4 addresses that are currently declined in a given subnet, so is a count of the number of leases currently unavailable. Once a lease is recovered, this statistic will be decreased. Ideally, this statistic should be zero. If this statistic is non-zero (or worse increasing), a network administrator should investigate if there is a misbehaving device in his network. The id is the subnet-id of a given subnet. This statistic is exposed for each subnet separately. reclaimed-declined-addresses integer This statistic shows the 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 This statistic shows the 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.

The management API allows the issuing of specific management commands, such as statistics retrieval, reconfiguration or shutdown. For more details, see . Currently the only supported communication channel type is 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 your 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 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: build-report config-get config-reload config-set config-test config-write dhcp-disable dhcp-enable leases-reclaim list-commands shutdown version-get as described in . In addition, it supports the following statistics related commands: statistic-get statistic-reset statistic-remove statistic-get-all statistic-reset-all statistic-remove-all as described here .

The following standards are currently supported: 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). DHCP Relay Agent Information Option, RFC 3046: Relay Agent Information option is supported. Vendor-Identifying Vendor Options for Dynamic Host Configuration Protocol version 4, RFC 3925: Vendor-Identifying Vendor Class and Vendor-Identifying Vendor-Specific Information options are supported. Client Identifier Option in DHCP Server Replies, RFC 6842: Server by default sends back client-id option. That capability may be disabled. See for details.

Kea allows loading hook libraries that sometimes could 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. User contexts can store arbitrary data as long as it is valid JSON syntax and its top level element is a map (i.e. the data must be enclosed in curly brackets). Some hook libraries may expect specific formatting, though. Please consult specific hook library documentation for details. User contexts can be specified on either global scope, shared network, subnet, pool, client class, option data or definition level, and host reservation. One other useful usage is the ability to store comments or descriptions. Let's consider an imaginary case of devices that have color 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 as maybe 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 specified in the user context as long as it is a valid JSON map. For example, the forementioned case of LED devices could be configured in the following way: "Dhcp4": { "subnet4": [ { "subnet": "192.0.2.0/24", "pools": [ { "pool": "192.0.2.10 - 192.0.2.20", // This is pool specific user context "user-context": { "colour": "red" } } ], // This is a subnet specific user context. You can put whatever type // of information you want as long as it is a valid JSON. "user-context": { "comment": "network on the second floor", "last-modified": "2017-09-04 13:32", "description": "you can put here anything you like", "phones": [ "x1234", "x2345" ], "devices-registered": 42, "billing": false } }, ... ], ... } It should be noted that Kea will not use that information, but will simply store and make it available to hook libraries. It is up to the hook library to extract that information and make use of it. The parser translates a "comment" entry into a user-context with the entry, this allows to attach a comment inside the configuration itself. For more background information, see .

These are the current limitations of the DHCPv4 server software. Most of them are reflections of the current stage of development and should be treated as not implemented yet, rather than actual limitations. However, some of them are implications of the design choices made. Those are clearly marked as such. BOOTP (RFC 951) is not supported. This is a design choice: BOOTP support is not planned. On Linux and BSD system families the DHCP messages are sent and received over the 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 the system kernel. Currently, Kea can only parse the data link layer headers with a format adhering to IEEE 802.3 standard and assumes this data link layer header format for all interfaces. Hence, Kea will fail to work on interfaces which use different data link layer header formats (e.g. Infiniband). The DHCPv4 server does not verify that assigned address is unused. According to RFC 2131, the allocating server should verify that address is not used by sending ICMP echo request.

A collection of simple to use examples for DHCPv4 component of Kea is available with the sources. It is located in doc/examples/kea4 directory. At the time of writing this text there were 15 examples, but the number is growing slowly with each release.