diff options
Diffstat (limited to 'doc/sphinx/arm/hooks-limits.rst')
-rw-r--r-- | doc/sphinx/arm/hooks-limits.rst | 199 |
1 files changed, 199 insertions, 0 deletions
diff --git a/doc/sphinx/arm/hooks-limits.rst b/doc/sphinx/arm/hooks-limits.rst new file mode 100644 index 0000000..25cf916 --- /dev/null +++ b/doc/sphinx/arm/hooks-limits.rst @@ -0,0 +1,199 @@ +.. _hooks-limits: + +``limits``: Limits to Manage Lease Allocation and Packet Processing +=================================================================== + +This hook library enables two types of limits: + +1. Lease limiting: allow a maximum of ``n`` leases assigned at any one time. +2. Rate limiting: allow a maximum of ``n`` packets per ``time_unit`` to receive a response. + +The Limits hook library is only available to ISC customers with a paid support contract. + +.. _hooks-limits-configuration: + +Configuration +~~~~~~~~~~~~~ + +The following examples are for ``kea-dhcp6``, but they apply equally to +``kea-dhcp4``. The wildcards ``"<limit-type>"`` and ``"<limit-value>"`` need to be replaced +with the respective keys and values for each limit type described in the sections following this +one. + +The library can be loaded by both ``kea-dhcp4`` and ``kea-dhcp6`` servers by adding its path in the +``"hooks-libraries"`` element of the server's configuration. + +.. code-block:: json + + { + "Dhcp6": { + "hooks-libraries": [ + { + "library": "/usr/local/lib/libdhcp_limits.so" + } + ] + } + } + +This alone does not limit anything. The desired limits are added to the user context in the +configuration portion of the element that identifies the clients to be limited: a client class or a +subnet. Upon reconfiguration, if Kea picked up on the configured limits, it logs one line for +each configured limit. The log message contains ``LIMITS_CONFIGURED`` in its identifier. + +This is how a lease limit is defined for a client class: + +.. code-block:: json + + { + "Dhcp6": { + "client-classes": [ + { + "name": "cable-modem-1", + "test": "option[123].hex == 0x000C4B1E", + "user-context": { + "limits": { + "<limit>": "<limit-value>" + } + } + } + ] + } + } + +This is how a lease limit is defined for a global subnet: + +.. code-block:: json + + { + "Dhcp6": { + "subnet6": [ + { + "id": 1, + "subnet": "2001:db8::/64", + "user-context": { + "limits": { + "<limit>": "<limit-value>" + } + } + } + ] + } + } + +This is how a lease limit is defined for a subnet inside a shared network: + +.. code-block:: json + + { + "Dhcp6": { + "shared-networks": [ + { + "subnet6": [ + { + "id": 1, + "subnet": "2001:db8::/64", + "user-context": { + "limits": { + "<limit>": "<limit-value>" + } + } + } + ] + } + ] + } + } + +.. note:: + + The Limits hook library uses the class name to identify a client class and the subnet ID to + identify a subnet. Changing a test expression in a client class or the network range of a + subnet while leaving the name or ID unchanged does not reset the lease count for the + respective client class or subnet. To reset the lease count, change the client class name + or the subnet ID. + +.. _hooks-limits-lease-limiting: + +Lease Limiting +~~~~~~~~~~~~~~ + +It is possible to limit the number of leases that a group of clients can get from a Kea DHCP server +or from a set of collaborating Kea DHCP servers. + +The value of a lease limit can be specified as an unsigned integer in 32 bits, i.e. between ``0`` and +``4,294,967,295``. Each lease type can be limited individually. IPv4 leases and IPv6 IA_NA leases +are limited through the ``"address-limit"`` configuration entry. IPv6 IA_PD leases are limited +through the ``"prefix-limit"`` configuration entry. Here are some examples: + +* ``"address-limit": 4`` +* ``"prefix-limit": 2`` + +For lease limiting, client classes and the associated lease counts - which are +checked against the configured limits - are updated for each lease in the following hook callouts: + +* ``lease4_select`` +* ``lease4_renew`` +* ``lease6_select`` +* ``lease6_renew`` +* ``lease6_rebind`` + +As a result, classes for which ``"only-if-required"`` is "true" cannot be lease-limited. +Please refer to :ref:`the classification steps <classify-classification-steps>` for more information on which +client classes can be used to limit the number of leases. + +.. note:: + + Under load, a Kea DHCP server may allocate more leases than the limit strictly allows. This only has a chance of + happening during high traffic surges, coming from clients belonging to the same class or the + same subnet, depending on what is limited. Users may be interested in following the development of + `atomic lease limits <https://gitlab.isc.org/isc-projects/kea/-/issues/2449>`__ in ISC's GitLab instance. + +.. _hooks-limits-rate-limiting: + +Rate Limiting +~~~~~~~~~~~~~ + +It is possible to limit the frequency or rate at which inbound packets receive a response. + +The value of a rate limit can be specified in the format ``"<p> packets per <time-unit>"``. ``<p>`` +is any number that can be represented by an unsigned integer in 32 bits, i.e. between ``0`` and +``4,294,967,295``. ``<time-unit>`` can be any of ``second``, ``minute``, ``hour``, ``day``, +``week``, ``month``, or ``year``. A ``month`` is considered to be 30 days for +simplicity; similarly, a ``year`` is 365 days for limiting purposes. This syntax +covers a wide range of rates, from one lease per year to four billion leases per +second. This value is assigned to the ``"rate-limit"`` configuration entry. +Here are some examples: + +* ``"rate-limit": 1 packet per second`` +* ``"rate-limit": 4 packets per minute`` +* ``"rate-limit": 16 packets per hour`` + +The configured value of ``0`` packets is a convenient way of disabling packet processing for certain +clients entirely. As such, it means its literal value and is not a special value for disabling +limiting altogether, as might be imagined. Disabling limiting entirely is achieved by removing +the ``"rate-limit"`` leaf configuration entry, the ``"limits"`` map or user context +around it, or the hook library configuration. The same applies to the value of ``0`` in lease +limiting. However, that use case is best achieved with rate limiting; it puts less computational +strain on Kea, since the action of dropping the request or sending a NAK is decided earlier. + +In terms of rate limiting, client classes are evaluated at the ``pkt4_receive`` and the +``pkt6_receive`` callout, respectively, so that rate limits are checked as early as possible in the +packet-processing cycle. Thus, only those classes which are assigned to the packet solely via an +independent test expression can be used. Classes that depend on host reservations or the special +``BOOTP`` or ``KNOWN`` classes, and classes that are marked with ``"only-if-required": true``, +cannot be rate limited. See :ref:`the classification steps <classify-classification-steps>` for +more details on which client classes can be used to limit the packet rate. + +Rate limits based on subnet are enforced only on the initially selected subnet for a given packet. +If the selected subnet is subsequently changed, as may be the case for subnets in a +shared network or when reselection is enabled in libraries such as the RADIUS hook, rate +limits on the newly selected subnet are ignored. In other words, packets are gated only by +the rate limit on the original subnet. + +.. note:: + + It may seem logical to think that assigning a rate limit of ``n`` packets per time unit results + in ``n`` DORA or ``n`` SARR exchanges. However, by default, all inbound packets are counted - meaning + that a full message exchange accounts for two packets. To achieve the effect of counting an + exchange only once, use client-class rate-limiting with a test expression that binds + ``pkt4.msgtype`` to DHCPDISCOVER messages or ``pkt6.msgtype`` to SOLICIT messages. |