Skip to content

Lease addresses with DHCP

A DHCPSubnet document makes the router lease addresses of one subnet on its links, IPv4 or IPv6 by the subnet's range. Kea's DHCPv4 and DHCPv6 servers answer the clients, from pools and from reservations by MAC address, and give them DNS servers and a domain. IPv4 clients also get a gateway and PXE boot files, and IPv6 routers on the link may get a prefix of their own. A DHCPRelay document makes the router relay its links' requests to DHCP servers elsewhere instead. This page serves an office LAN, reserves an address, boots PXE clients, serves a VRF's links, serves IPv6, delegates prefixes, relays requests and checks the leases. DHCP on a router explains how chalkos runs Kea and dnsmasq.

Before you begin

You need a router set up as Set up a router describes, and the secrets file. Its image must carry Kea. Router images do unless their modules set the image option chalkos.router.dhcp.enable to false. An image without Kea does not accept DHCPSubnet documents: chalkctl validate reports that it does not accept kind "DHCPSubnet". Ask the image author for an image with DHCP; Leave router features out lists the options of every router feature.

Relays need dnsmasq, which router images carry unless their modules set chalkos.router.dhcpRelay.enable to false. An image without it refuses DHCPRelay documents the same way.

The examples configure a LAN router r1 with these links:

  • lan0, the office LAN, with the addresses 192.0.2.1/24 and 2001:db8:1::1/64.
  • vlan20, a VLAN on lan0 for guests, in the VRF guest, with the address 198.51.100.1/24.
  • cust0, a link to customers' home routers, in the VRF access, with the address 2001:db8:2::1/64.

Each link the router serves needs a Network document that names it in [Match] Name= and gives it a static address in the subnet. Kea answers a link's clients from that address.

Serve a LAN

The DHCPSubnet office leases addresses of 192.0.2.100 to 192.0.2.199 on lan0:

apiVersion: router.chalkos.dev/v1alpha1
kind: DHCPSubnet
metadata: {name: office}
spec:
  interfaces: [lan0]
  subnet: 192.0.2.0/24
  pools: [{first: 192.0.2.100, last: 192.0.2.199}]
  gateway: 192.0.2.1
  dns: [192.0.2.1]
  domain: lan.example.org
---
apiVersion: networkd.chalkos.dev/v1alpha1
kind: Network
metadata: {name: 10-lan0}
spec:
  match: {name: lan0}
  network: {address: [192.0.2.1/24], vlan: [vlan20], ipv6AcceptRA: false}

The clients route through gateway and ask the DNS servers of dns, here the router itself. A Resolver on 192.0.2.1 answers them, as Translate to IPv4 and resolve names describes. domain is the domain of the clients' names, which they also search. A lease lasts 12 hours unless leaseTime says otherwise, such as 1h or 7d.

interfaces names each link by its full name; it takes no glob such as lan*. Kea reads and answers DHCPv4 through raw packet sockets on exactly these links, and nftables does not see that traffic. The router's firewall opens no port for it and a FirewallPolicy cannot filter it, so the list of names decides which links the server answers on. A link takes one subnet of each family.

Reserve addresses

A reservation gives a client the same address every time, by its MAC address:

spec:
  interfaces: [lan0]
  subnet: 192.0.2.0/24
  pools: [{first: 192.0.2.100, last: 192.0.2.199}]
  reservations:
    - {mac: "52:54:00:12:34:56", address: 192.0.2.10, hostname: printer}

Write the MAC address in lower case. A reserved address lies inside the subnet and outside its pools. Neither a pool nor a reservation may hold the subnet's network or broadcast address, the router's own address on the link, the gateway, a DNS server or the boot server. hostname is the name the client gets; leave it out for none. A subnet without pools leases its reservations alone.

Boot PXE clients

boot gives PXE clients the file to load, by their firmware. The files are served elsewhere, by a TFTP or HTTP server:

spec:
  interfaces: [lan0]
  subnet: 192.0.2.0/24
  pools: [{first: 192.0.2.100, last: 192.0.2.199}]
  boot:
    server: 192.0.2.5
    bios: undionly.kpxe
    uefi: ipxe.efi
    ipxe: http://192.0.2.5/boot.ipxe

A client with BIOS firmware, of architecture 0 in its option 93, gets bios. One with x86-64 UEFI firmware, of architecture 7 or 9, gets uefi. Both load the file from the TFTP server at server, so bios and uefi need server. A client that already runs iPXE says so in its user class, option 77, and gets ipxe instead, here a script over HTTP. Without ipxe, iPXE would get its own binary again and load it in a loop. Give the files that your clients need; a file left out goes to no client. A boot with ipxe alone, an HTTP URL, needs no server.

A subnet whose links are in a VRF is served by a DHCP server of that VRF, apart from the default VRF's. The guests' subnet on vlan20, in the VRF guest, leases for an hour:

apiVersion: router.chalkos.dev/v1alpha1
kind: DHCPSubnet
metadata: {name: guests}
spec:
  interfaces: [vlan20]
  subnet: 198.51.100.0/24
  pools: [{first: 198.51.100.10, last: 198.51.100.250}]
  gateway: 198.51.100.1
  dns: [198.51.100.1]
  leaseTime: 1h
---
apiVersion: router.chalkos.dev/v1alpha1
kind: VRF
metadata: {name: guest}
spec: {table: 300, interfaces: [vlan20]}
---
apiVersion: networkd.chalkos.dev/v1alpha1
kind: NetDev
metadata: {name: 20-vlan20}
spec:
  netDev: {kind: vlan, name: vlan20}
  vlan: {id: 20}
---
apiVersion: networkd.chalkos.dev/v1alpha1
kind: Network
metadata: {name: 20-vlan20}
spec:
  match: {name: vlan20}
  network: {address: [198.51.100.1/24], ipv6AcceptRA: false}

A subnet's links are all in one VRF. Two subnets of one VRF do not overlap, while subnets of different VRFs may use the same range. Separate routing with VRFs declares the VRF.

Serve IPv6

A subnet of an IPv6 range makes Kea's DHCPv6 server lease its addresses. The DHCPSubnet office6 leases 2001:db8:1::100 to 2001:db8:1::1ff on lan0 and reserves an address for the printer:

apiVersion: router.chalkos.dev/v1alpha1
kind: DHCPSubnet
metadata: {name: office6}
spec:
  interfaces: [lan0]
  subnet: 2001:db8:1::/64
  pools: [{first: "2001:db8:1::100", last: "2001:db8:1::1ff"}]
  dns: ["2001:db8:1::1"]
  domain: lan.example.org
  reservations:
    - {mac: "52:54:00:12:34:56", address: "2001:db8:1::10", hostname: printer}
---
apiVersion: networkd.chalkos.dev/v1alpha1
kind: Network
metadata: {name: 10-lan0}
spec:
  match: {name: lan0}
  network: {address: [192.0.2.1/24, "2001:db8:1::1/64"], vlan: [vlan20], ipv6AcceptRA: false, ipv6SendRA: true}
  ipv6SendRA: {managed: true}
  ipv6Prefix: {prefix: "2001:db8:1::/64"}

IPv6 clients ask a DHCPv6 server only when the router advertisements of their link tell them to. They also learn their routers from the advertisements. The link's Network sends them, with ipv6SendRA: true in network, and its ipv6SendRA section sets the flags. A subnet that leases addresses, from pools or reservations, needs managed: true. One that gives only options or prefixes needs managed: true or otherInformation: true. The validator router refuses a subnet whose link's advertisements lack the flag. It also reads networkd's former names of these keys, ipv6PrefixDelegation in network and the section ipv6PrefixDelegation.

An IPv6 subnet differs from an IPv4 one in these ways:

  • It is served on one link: interfaces names one link, as Kea's DHCPv6 server picks a subnet by its link.
  • It has no gateway and no boot. dns lists IPv6 addresses.
  • A pool or reservation does not hold the subnet's first address, the subnet-router anycast address.
  • Kea learns a client's MAC address from its DUID, when it is a DUID-LL or DUID-LLT, or from an EUI-64 link-local address. A client with another DUID and a random link-local address does not match a reservation by MAC address and takes an address of the pools.
  • Clients prefer their addresses for the whole leaseTime and renew at half of it.

A link takes one subnet of each family, so lan0 serves office and office6 side by side.

Delegate prefixes to other routers

prefixDelegation gives routers on the link a prefix each to number their own networks, as a provider gives its customers' home routers. The subnet customers delegates a /56 of 2001:db8:100::/40 to each home router on cust0 that asks, for a day:

apiVersion: router.chalkos.dev/v1alpha1
kind: DHCPSubnet
metadata: {name: customers}
spec:
  interfaces: [cust0]
  subnet: 2001:db8:2::/64
  dns: ["2001:db8:2::1"]
  leaseTime: 1d
  prefixDelegation: {prefix: "2001:db8:100::/40", length: 56}
---
apiVersion: networkd.chalkos.dev/v1alpha1
kind: Network
metadata: {name: 20-cust0}
spec:
  match: {name: cust0}
  network: {address: ["2001:db8:2::1/64"], ipv6SendRA: true, ipv6AcceptRA: false}
  ipv6SendRA: {otherInformation: true}

length is longer than the range's prefix: here the range holds 65 536 prefixes of /56. In the link's VRF, the range overlaps no subnet and no other delegated range. It holds none of the router's own addresses, no destination of a Network's routes and no NAT64's prefix.

The router routes each delegated prefix to the link-local address of the router that asked for it, in the link's VRF, until its lease ends. A release or the end of the lease removes the route, and the routes survive the router's boot.

Relay requests to another server

A DHCPRelay forwards the DHCP requests of its links' clients to servers elsewhere, through the links' VRF, and their answers back. The relay office relays lan0 to a central server of each family behind the uplink wan0:

apiVersion: router.chalkos.dev/v1alpha1
kind: DHCPRelay
metadata: {name: office}
spec:
  interfaces: [lan0]
  servers: [203.0.113.53, "2001:db8:ff::53"]

DHCPv4 requests go to the IPv4 servers and DHCPv6 requests to the IPv6 ones. The relay forwards a link's requests from the router's address of the family on the link. The server picks the client's subnet by that address, so the link's Network gives the link a static address of each family the servers have, a global one for IPv6. The servers are unicast addresses. A DHCPv6 server answers relayed requests on such an address, as Kea's does on an interface given as <name>/<address>, such as "interfaces": ["eth0/2001:db8:ff::53"].

A relay in a VRF reaches its servers through the VRF's routes. The relay guests relays DHCPv4 of vlan20, in the VRF guest, to a server on the VRF's own uplink vlan30, a VLAN on wan0:

apiVersion: router.chalkos.dev/v1alpha1
kind: DHCPRelay
metadata: {name: guests}
spec:
  interfaces: [vlan20]
  servers: [203.0.113.154]
---
apiVersion: router.chalkos.dev/v1alpha1
kind: VRF
metadata: {name: guest}
spec: {table: 300, interfaces: [vlan20, vlan30]}
---
apiVersion: networkd.chalkos.dev/v1alpha1
kind: Network
metadata: {name: 31-vlan30}
spec:
  match: {name: vlan30}
  network: {address: [203.0.113.130/25], ipv6AcceptRA: false}

vlan30 also needs a NetDev and a place in wan0's vlan list, as vlan20 has on lan0. The example dhcp-relay.yaml holds the whole router.

A relay's links are in one VRF, each link takes one relay, and interfaces names each link, without a glob, as a subnet's does. Relays of one family may run in any VRFs, the default one included, as office and guests relay DHCPv4 here. A family that the router relays it does not serve in the same VRF, nor when either the relay or the server is in the default VRF. Kea's and dnsmasq's sockets of a family share a port, which works only when each is bound to a VRF of its own. Relay DHCPv4 from a VRF and serve DHCPv4 in another VRF, or relay one family and serve the other anywhere.

Check that it worked

Deliver the documents as any config:

chalkctl validate
chalkctl config apply r1 --secrets secrets.age

chalkctl status shows each VRF's DHCP server and, while it runs, the addresses each subnet leased and the addresses of its pools. A reserved address counts as leased once its client took it:

chalkctl status r1 --secrets secrets.age
dhcp4 default: running
  SUBNET  RANGE         LEASED  POOL
  office  192.0.2.0/24  12      100
dhcp6 default: running
  SUBNET   RANGE            LEASED  POOL  DELEGATED  DELEGABLE
  office6  2001:db8:1::/64  9       256   0          0
dhcp4 guest: running
  SUBNET  RANGE            LEASED  POOL
  guests  198.51.100.0/24  3       241
dhcp6 access: running
  SUBNET     RANGE            LEASED  POOL  DELEGATED  DELEGABLE
  customers  2001:db8:2::/64  0       0     2          65536

A DHCPv6 server's subnets also count the prefixes they delegated and those they may delegate. A count past 18446744073709551615, as of the addresses of a /64 pool, shows as >=2^64. A router that relays shows a line for each VRF's relay, such as dhcp-relay guest: running.

A server or relay that does not run says why, such as dhcp4 guest: not running: kea-dhcp4@guest.service is failed. chalkctl logs r1 --unit kea-dhcp6@access.service shows Kea's log, chalkctl logs r1 --unit chalkos-dhcp.service the log of the unit that starts the servers and chalkctl logs r1 --unit dnsmasq-relay@guest.service a relay's.

The routes of delegated prefixes have the protocol dhcp-pd:

chalkctl node routes r1 --secrets secrets.age --vrf access --protocol dhcp-pd
DESTINATION            TYPE     PROTOCOL  TABLE  METRIC  NEXT HOPS
2001:db8:100::/56      unicast  dhcp-pd   500    1024    fe80::5054:ff:fe12:3456 on cust0
2001:db8:100:100::/56  unicast  dhcp-pd   500    1024    fe80::5054:ff:fe65:4321 on cust0

The diagnostic dhcp-leases lists the leases, a line each with the VRF, the subnet, the address or the delegated prefix and its length, the MAC address, or - when the server learned none, the host name and when the lease ends. subnet and mac narrow the list. It takes the operator role, as the leases show which machines are on the networks:

chalkctl node diag r1 --secrets secrets.age dhcp-leases subnet=office
VRF SUBNET  ADDRESS MAC HOSTNAME    ENDS
default office  192.0.2.10  52:54:00:12:34:56   printer 2026-11-01T22:00:00Z
default office  192.0.2.100 52:54:00:aa:bb:cc   -   2026-11-01T21:47:12Z

Fix a refused delivery

The schema of DHCPSubnet takes each interface by its name, so chalkctl validate refuses a glob before chalkctl sends anything: DHCPSubnet "office": - at '/spec/interfaces/0': 'lan*' does not match pattern '^(?:[A-Za-z0-9_.:@-]{1,15})$'. Name each link instead.

The rules of DHCPSubnet and DHCPRelay refuse documents at the same point, and so does the check of the values the kinds take once, such as a reservation's MAC address. A subnet's rules see one document at a time, so chalkctl validate checks its pools and reservations against each other as well. The built-ins' reference lists each rule with its expression:

Message Fix
DHCPSubnet/office: spec.subnet: the subnet 192.0.2.0/31 is no IPv4 range of 30 bits at most or IPv6 range of 120 bits at most of addresses a DHCP server leases Give an IPv4 range of 30 bits at most, apart from 0.0.0.0/8, 127.0.0.0/8, 169.254.0.0/16 and 224.0.0.0/3, or an IPv6 range of 120 bits at most, apart from ::/8, fe80::/10 and ff00::/8.
DHCPSubnet/office: spec: the pool 192.0.2.100-192.0.2.199 lies outside the subnet 192.0.2.0/25 Move the pool into the subnet, or widen the subnet.
DHCPSubnet/office: spec: the pool 192.0.2.199-192.0.2.100 ends before it starts Swap first and last.
DHCPSubnet/office: spec: the pool 192.0.2.100-192.0.2.150 shares addresses with another pool Give each pool addresses of its own.
DHCPSubnet/office: spec: the pool 192.0.2.0-192.0.2.10 holds 192.0.2.0, the network address of the subnet 192.0.2.0/24 Start the pool after the subnet's first address and end it before its last.
DHCPSubnet/office6: spec: the pool 2001:db8:1::-2001:db8:1::ff holds 2001:db8:1::, the subnet-router anycast address of the subnet 2001:db8:1::/64 Start the pool after the subnet's first address.
DHCPSubnet/office: spec: the pool 192.0.2.100-192.0.2.199 holds 192.0.2.150, the gateway Keep the gateway, the DNS servers and boot's server out of the pools and reservations.
DHCPSubnet/office: spec: the reservation of 52:54:00:12:34:56, 198.51.100.10, lies outside the subnet 192.0.2.0/24 Reserve an address of the subnet.
DHCPSubnet/office: spec: the reservation of 52:54:00:12:34:56, 192.0.2.120, lies in the pool 192.0.2.100-192.0.2.199; reserve addresses outside the pools Reserve an address outside the pools, or shrink the pool.
DHCPSubnet/office: spec: the reservation of 52:54:00:12:34:56, 192.0.2.255, is the broadcast address of the subnet 192.0.2.0/24 Reserve an address between the subnet's first and last address.
DHCPSubnet/office: spec: the reservation of 52:54:00:12:34:56, 192.0.2.53, is a DNS server Keep the gateway, the DNS servers and boot's server out of the pools and reservations.
DHCPSubnet/office: spec: the gateway 198.51.100.1 lies outside the subnet 192.0.2.0/24 Give a gateway of the subnet.
DHCPSubnet/office: spec: the gateway, 192.0.2.0, is the network address of the subnet 192.0.2.0/24 Give a gateway between the subnet's first and last address.
DHCPSubnet/office: spec: the DNS server 2001:db8::53 is no IPv4 address Give DNS servers of the subnet's family.
DHCPSubnet/office6: spec: gateway is for IPv4 subnets; IPv6 clients learn their routers from router advertisements Drop gateway; send router advertisements from the link's Network.
DHCPSubnet/office6: spec: boot is for IPv4 subnets Drop boot.
DHCPSubnet/office: spec: prefixDelegation delegates IPv6 prefixes, which an IPv4 subnet has none of Drop prefixDelegation, or delegate in an IPv6 subnet.
DHCPSubnet/office6: spec: an IPv6 subnet is served on one link, the one Kea's subnet names; give each of lan0, lan1 a DHCPSubnet of its own Name one link.
DHCPSubnet/customers: spec.prefixDelegation: prefixDelegation's prefix 2001:db8:100::/56 is no IPv6 range shorter than 56 bits apart from ::/8, fe80::/10 and ff00::/8 Give an IPv6 range apart from those ranges and a longer length.
DHCPSubnet/customers: spec: prefixDelegation's prefix 2001:db8::/32 overlaps the subnet 2001:db8:2::/64 Delegate from a range apart from the subnet.
DHCPSubnet/office: spec.leaseTime: a lease of 999999d is longer than 4294967295 seconds, the longest Kea takes Shorten the lease to 49710 days or less.
DHCPSubnet/office: spec.boot: boot names no file; give bios, uefi or ipxe, or leave boot out Give a file, or drop boot.
DHCPSubnet/office: spec.boot: boot's server is no IPv4 address Give the TFTP server's IPv4 address.
DHCPSubnet/office: spec.boot: boot's bios and uefi files load over TFTP from boot's server; give server Give the TFTP server's address in server.
DHCPSubnet/office: spec.interfaces[0], spec.interfaces[2]: "lan0" is not unique; name each link once Name each link once.
DHCPSubnet/office: spec.reservations[0].mac, spec.reservations[2].mac: "52:54:00:12:34:56" is not unique; keep one reservation of the MAC address Keep one reservation of the MAC address.
DHCPSubnet/office: spec.reservations[1].address, spec.reservations[3].address: "192.0.2.10" is not unique; reserve each address once Reserve each address once.
DHCPRelay/office: spec.servers: the server ff05::1:3 is no unicast address a relay forwards to Name the servers by their unicast addresses.
DHCPRelay/office: spec.servers[0], spec.servers[1]: "2001:db8:ff::53" is not unique; name each server once Name each server once.
DHCPRelay/guests: spec.interfaces[0], DHCPRelay/office: spec.interfaces[1]: "lan0" is not unique; a link takes one relay Relay the link in one document, and name it once there.

The validator router checks what spans documents: the links' VRFs and addresses, which Network documents give, the router advertisements of an IPv6 subnet's link, and the subnets and relays of a VRF together. chalkctl validate and chalkctl config apply run it before chalkctl sends anything, and the router runs it again at the delivery. It checks the VRFs' links first, and the subnets and relays only once the links pass. Within each step it reports every fault it finds at once. chalkctl prints the refusals after the config is refused:, as it prints a rule's, each with the document and the field it is about:

Message Fix
DHCPSubnet/office: spec.pools[0]: the pool 192.0.2.10-192.0.2.50 holds 192.0.2.30, the router's own address on the subnet's links Keep the router's addresses on the links out of the pools and reservations.
DHCPSubnet/office: spec.interfaces[0]: Network 10-lan0 gives lan0 no static address in the subnet 203.0.113.0/24, which Kea answers its clients from Give the link's Network an address in the subnet.
DHCPSubnet/office: spec.interfaces[0]: no Network document matches lan9 by Name= alone, so the router cannot find its address in the subnet 192.0.2.0/24; give the link one Add a Network that names the link.
DHCPSubnet/office: spec.interfaces[1]: its links are in VRF default and VRF guest; a subnet is served in one VRF Give each VRF's links a subnet of their own.
DHCPSubnet/office: spec.interfaces: its links are in the VRF "lan", which no VRF document declares Declare the VRF with a VRF document, or drop VRF= from the link's Network.
DHCPSubnet/office: spec.interfaces[0]: guest is the device of VRF guest; name the VRF's links Name the VRF's links.
DHCPSubnet/office: spec.interfaces[0]: n64-edge is the link of NAT64 edge Leave the NAT64's link out.
DHCPSubnet/office: spec.interfaces[0]: DHCPSubnet lab and DHCPSubnet office both serve lan0; a link takes one subnet of each family Serve the link in one subnet of each family.
DHCPSubnet/servers: spec.subnet: DHCPSubnet office and DHCPSubnet servers overlap in VRF default: 192.0.2.0/24 and 192.0.2.64/26 Give the subnets ranges apart.
DHCPSubnet/office6: spec.interfaces[0]: Network 10-lan0 sends no router advertisements on lan0 with Managed=yes, which tell its clients to ask the DHCPv6 server for addresses; give it ipv6SendRA: true in network and managed: true in ipv6SendRA Send router advertisements with the Managed flag from the link's Network.
DHCPSubnet/customers: spec.interfaces[0]: Network 20-cust0 sends no router advertisements on cust0 with Managed=yes or OtherInformation=yes, which tell its clients to ask the DHCPv6 server; give it ipv6SendRA: true in network and otherInformation: true in ipv6SendRA Send router advertisements with the OtherInformation flag.
DHCPSubnet/customers: spec.prefixDelegation.prefix: prefixDelegation's prefix 2001:db8:100::/40 holds 2001:db8:100::1, the address Network 30-lan3 gives a link of VRF access Delegate from a range that holds none of the router's addresses in the VRF.
DHCPSubnet/customers: spec.prefixDelegation.prefix: prefixDelegation's prefix 2001:db8:100::/40 holds the route of 2001:db8:100:100::/56 that Network 30-lan3 gives a link of VRF access Delegate from a range apart from the VRF's static routes.
DHCPSubnet/customers: spec.prefixDelegation.prefix: prefixDelegation's prefix 2001:db8:100::/40 holds NAT64 edge's prefix 2001:db8:100:ff::/96 of VRF access Delegate from a range apart from the NAT64's prefix.
DHCPSubnet/customers: spec.prefixDelegation.length: the DHCPSubnet documents delegate more than 65536 prefixes, the most the router routes; delegate fewer Delegate from smaller ranges or delegate shorter prefixes, so that all subnets together delegate 65 536 prefixes at most. The refusal names the first subnet, by name, whose prefixes cross the limit.
DHCPRelay/office: spec.interfaces[0]: Network 10-lan0 gives lan0 no static global IPv6 address, which the relay forwards its clients' requests from Give the link a static address of each family the relay's servers have.
DHCPSubnet/lab: spec.interfaces: DHCPRelay office relays DHCPv4 in VRF default and DHCPSubnet lab serves DHCPv4 in VRF lab; dnsmasq's and Kea's sockets of one family bind beside each other only in two VRFs other than the default one Relay and serve the family in two VRFs other than the default one.

The renderer of the output kea refuses one more config when the router renders it, at a delivery or at chalkctl config apply --dry-run, and chalkctl validate does not run it:

Message Fix
DHCPSubnet/office6: spec.subnet: DHCPSubnet lab6 and DHCPSubnet office6 get the same number 2387838497 in Kea; change one subnet's range Change the range of one IPv6 subnet, whose number is a hash of its range.

Limits

  • Kea reads and answers DHCPv4 on raw sockets, ahead of the router's firewall. A FirewallPolicy does not filter DHCPv4 on a served link, and the firewall opens no port for it. DHCPv6 and the relays take UDP sockets, and the firewall opens their ports on their links alone.
  • The router serves its own links. Its servers do not answer requests that another relay forwards from other networks, and a prefix delegated through a relay is not routed.
  • A family relayed and served on one router needs two VRFs other than the default one.
  • DHCPv6 reservations match a client's MAC address only when its DUID or link-local address holds it.
  • A link takes its address from a Network document. The validator router cannot check an address that a link takes from DHCP or that another program sets, so such a link is refused.
  • The leases live in a file of each server on the node's /var. Two routers do not share them, and a reinstall that wipes /var forgets them and the routes of their prefixes.
  • Kea tells the router of a release without waiting for it to record the lease before. A release right after its lease can leave the prefix routed until the lease would have ended, the subnet's leaseTime at most.
  • The router routes 65 536 delegated prefixes at most, of all its subnets together. The validator router refuses subnets that delegate more, such as two ranges of /40 that delegate /56s.
  • A DHCPSubnet holds 32 pools, 512 reservations and 16 DNS servers at most, and the schema refuses longer lists. The subnet's rules compare each reservation with each pool, and each pool and reservation with the servers. The bounds keep these checks within the cost a node allows a rule.
  • The servers update no DNS records for their clients.

What next