UNBOUND.CONF(5)

💖 `unbound.conf(5)` — the Unbound 1.26.2 config file, bestie edition 🎀

✨ NAME ✨

unbound.conf — Unbound 1.26.2 configuration file.

🌸 SYNOPSIS 🌸

unbound.conf

🦋 DESCRIPTION 🦋

unbound.conf is used to configure unbound(8).

The utility unbound-checkconf(8) can be used to check unbound.conf prior to usage.

💅 FILE FORMAT 💅

Whitespace is used to separate keywords. Whitespace indentation is insignificant, but it is still recommended for visual clarity. Comments start with # and last to the end of line. Empty lines are ignored, as is whitespace at the beginning of a line.

Attribute keywords end with a colon (:) and they are either options or section clauses — section clauses group options together.

The configuration file is logically divided into sections, where each section is introduced by a section clause.

🌈 EXAMPLE 🌈

An example minimal config file is shown below; most settings are the defaults. Copy this to /etc/unbound/unbound.conf and start the server with:

$ unbound -c /etc/unbound/unbound.conf

Stop the server with:

$ kill `cat /run/unbound.pid`

The source distribution contains an extensive example.conf file with all the options.

# unbound.conf(5) config file for unbound(8).
server:
    directory: "/etc/unbound"
    username: unbound
    # make sure unbound can access entropy from inside the chroot.
    # e.g. on linux the use these commands (on BSD, devfs(8) is used):
    #      mount --bind -n /dev/urandom /etc/unbound/dev/urandom
    # and  mount --bind -n /dev/log /etc/unbound/dev/log
    chroot: "/etc/unbound"
    # logfile: "/etc/unbound/unbound.log"  #uncomment to use logfile.
    pidfile: "/run/unbound.pid"
    # verbosity: 1      # uncomment and increase to get more logging.
    # listen on all interfaces, answer queries from the local subnet.
    interface: 0.0.0.0
    interface: ::0
    access-control: 10.0.0.0/8 allow
    access-control: 2001:db8::/64 allow

🎀 SECTION CLAUSES 🎀

The recognized section clauses are:

server: — Most of the configuration is found in this section.
remote-control: — Configuration for the facility used by unbound-control(8).
stub-zone: — Configuration for a zone that redirects to specific authoritative name servers, e.g. for zones not generally available on the greater Internet.
forward-zone: — Configuration for a zone that forwards to specific DNS resolvers.
auth-zone: — Configuration for local authoritative zones.
view: — Overriding a small subset of configuration for incoming requests. Requests are mapped to views with access-control-view and interface-view.
python: — Configuration for the optional python script module.
dynlib: — Configuration for the optional dynlib module that loads dynamic libraries into Unbound.
dnscrypt: — Configuration for the optional DNSCrypt feature.
cachedb: — Configuration for the optional cachedb module that can interface with second level caches, currently Redis or Redis-complatible databases.
dnstap: — Configuration of the optional dnstap logging feature; a flexible, structured binary log format for DNS software.
rpz: — Configuration for Response Policy Zones that allows for DNS filtering. Requires the respip module.

Section clauses can be repeated throughout the file (or included files) to logically group options in one visually cohesive group. This may be particularly useful for the server: clause with its myriad of options.

🧚 INCLUDING FILES 🧚

Files can be included using the include: directive. It can appear anywhere, and it accepts a single file name as argument. Processing continues as if the text from the included file was copied into the config file at that point. If also using chroot, using full path names for the included files works; relative pathnames for the included names work if the directory where the daemon is started equals its chroot/working directory or is specified before the include statement with directory: dir. Wildcards can be used to include multiple files, see glob(7).

For a more structural include option, the include-toplevel: directive can be used. This closes whatever section clause is currently active (if any) and forces the use of section clauses in the included files and right after this directive.

👑 SERVER OPTIONS 👑

These options are part of the server: section.

verbosity: <number> — The verbosity level.
- Level 0 — No verbosity, only errors.
- Level 1 — Gives operational information.
- Level 2 — Gives detailed operational information including short information per query.
- Level 3 — Gives query level information, output per query.
- Level 4 — Gives algorithm level information.
- Level 5 — Logs client identification for cache misses.

The verbosity can also be increased from the command line and during run time via remote control. See unbound(8) and unbound-control(8) respectively.

Default: 1

statistics-interval: <seconds> — The number of seconds between printing statistics to the log for every thread. Disable with value 0 or "". The histogram statistics are only printed if replies were sent during the statistics interval; requestlist statistics are printed for every interval (but can be 0). This is because the median calculation requires data to be present.

Default: 0 (disabled)

statistics-cumulative: <yes or no> — If enabled, statistics are cumulative since starting Unbound, without clearing the statistics counters after logging the statistics.

Default: no

extended-statistics: <yes or no> — If enabled, extended statistics are printed from unbound-control(8). The counters are listed in unbound-control(8). Keeping track of more statistics takes time.

Default: no

statistics-inhibit-zero: <yes or no> — If enabled, selected extended statistics with a value of 0 are inhibited from printing with unbound-control(8). These are query types, query classes, query opcodes, answer rcodes (except NOERROR, FORMERR, SERVFAIL, NXDOMAIN, NOTIMPL, REFUSED) and PRZ actions.

Default: yes

num-threads: <number> — The number of threads to create to serve clients. Use 1 for no threading.

Default: 1

port: <port number> — The port number on which the server responds to queries.

Default: 53

interface: <IP address or interface name[@port]> — Interface to use to connect to the network. This interface is listened to for queries from clients, and answers to clients are given from it. Can be given multiple times to work on several interfaces. If none are given the default is to listen on localhost.

If an interface name is used instead of an IP address, the list of IP addresses on that interface are used. The interfaces are not changed on a reload (kill -HUP) but only on restart.

A port number can be specified with @port (without spaces between interface and port number); if not specified the default port (from port) is used.

ip-address: <IP address or interface name[@port]> — Same as interface (for ease of compatibility with nsd.conf(5)).
interface-automatic: <yes or no> — Listen on all addresses on all (current and future) interfaces, detect the source interface on UDP queries and copy them to replies. This is a lot like ip-transparent, but this option services all interfaces whilst with ip-transparent you can select which (future) interfaces Unbound provides service on. This feature is experimental, and needs support in your OS for particular socket options.

Default: no

interface-automatic-ports: "<string>" — List the port numbers that interface-automatic listens on. If empty, the default port is listened on. The port numbers are separated by spaces in the string.

This can be used to have interface automatic deal with the interface, and listen on the normal port number, by including it in the list, and also HTTPS or DNS-over-TLS port numbers by putting them in the list as well.

Default: ""

outgoing-interface: <IPv4/IPv6 address or IPv6 netblock> — Interface to use to connect to the network. This interface is used to send queries to authoritative servers and receive their replies. Can be given multiple times to work on several interfaces. If none are given the default (all) is used. You can specify the same interfaces in interface and outgoing-interface lines; the interfaces are then used for both purposes. Outgoing queries are sent via a random outgoing interface to counter spoofing.

If an IPv6 netblock is specified instead of an individual IPv6 address, outgoing UDP queries will use a randomised source address taken from the netblock to counter spoofing. Requires the IPv6 netblock to be routed to the host running Unbound, and requires OS support for unprivileged non-local binds (currently only supported on Linux). Several netblocks may be specified with multiple outgoing-interface options, but do not specify both an individual IPv6 address and an IPv6 netblock, or the randomisation will be compromised. Consider combining with prefer-ip6: yes to increase the likelihood of IPv6 nameservers being selected for queries. On Linux you need these two commands to be able to use the freebind socket option to receive traffic for the ip6 netblock:

ip -6 addr add mynetblock/64 dev lo && \
ip -6 route add local mynetblock/64 dev lo

outgoing-range: <number> — okay bestie, this is the number of ports to open. This many file descriptors can be opened per thread. Must be at least 1. Default depends on compile options. Larger numbers need extra resources from the operating system. For performance a very large value is best, use libevent to make this possible. Should be higher (preferably double) than the value of num-queries-per-thread to account for cases where the request list is full and avoid file descriptor starvation. Default: 4096 (libevent) / 960 (minievent) / 48 (windows) 💅
outgoing-port-permit: <port number or range> — Permit Unbound to open this port or range of ports for use to send queries. A larger number of permitted outgoing ports increases resilience against spoofing attempts. Make sure these ports are not needed by other daemons. By default only ports above 1024 that have not been assigned by IANA are used. Give a port number or a range of the form low-high, without spaces. the permit and avoid statements are processed in the line order of the config file, adding the permitted ports and subtracting the avoided ports from the set of allowed ports. The processing starts with the non IANA allocated ports above 1024 in the set of allowed ports. ✨
outgoing-port-avoid: <port number or range> — Do not permit Unbound to open this port or range of ports for use to send queries. Use this to make sure Unbound does not grab a port that another daemon needs. The port is avoided on all outgoing interfaces, both IPv4 and IPv6. By default only ports above 1024 that have not been assigned by IANA are used. Give a port number or a range of the form low-high, without spaces. 🫶
outgoing-num-tcp: <number> — Number of outgoing TCP buffers to allocate per thread. If set to 0, or if do-tcp: no is set, no TCP queries to authoritative servers are done. For larger installations increasing this value is a good idea. Default: 10 🔥
incoming-num-tcp: <number> — Number of incoming TCP buffers to allocate per thread. If set to 0, or if do-tcp: no is set, no TCP queries from clients are accepted. For larger installations increasing this value is a good idea. Default: 10 🌸
edns-buffer-size: <number> — Number of bytes size to advertise as the EDNS reassembly buffer size. This is the value put into datagrams over UDP towards peers. The actual buffer size is determined by msg-buffer-size (both for TCP and UDP). Do not set higher than that value. Setting to 512 bypasses even the most stringent path MTU problems, but is seen as extreme, since the amount of TCP fallback generated is excessive (probably also for this resolver, consider tuning outgoing-num-tcp). Default: 1232 (DNS Flag Day 2020 recommendation) 👑
max-udp-size: <number> — Maximum UDP response size (not applied to TCP response). 65536 disables the UDP response size maximum, and uses the choice from the client, always. Suggested values are 512 to 4096. Default: 1232 (same as edns-buffer-size) 🦋
stream-wait-size: <number> — Number of bytes size maximum to use for waiting stream buffers. A plain number is in bytes, append 'k', 'm' or 'g' for kilobytes, megabytes or gigabytes (1024*1024 bytes in a megabyte). As TCP and TLS streams queue up multiple results, the amount of memory used for these buffers does not exceed this number, otherwise the responses are dropped. This manages the total memory usage of the server (under heavy use), the number of requests that can be queued up per connection is also limited, with further requests waiting in TCP buffers. Default: 4m 💖
msg-buffer-size: <number> — Number of bytes size of the message buffers. Default is 65552 bytes, enough for 64 Kb packets, the maximum DNS message size. No message larger than this can be sent or received. Can be reduced to use less memory, but some requests for DNS data, such as for huge resource records, will result in a SERVFAIL reply to the client. Default: 65552 😭
msg-cache-size: <number> — Number of bytes size of the message cache. A plain number is in bytes, append 'k', 'm' or 'g' for kilobytes, megabytes or gigabytes (1024*1024 bytes in a megabyte). Default: 4m 🎀
msg-cache-slabs: <number> — Number of slabs in the message cache. Slabs reduce lock contention by threads. Must be set to a power of 2. Setting (close) to the number of cpus is a fairly good setting. If left unconfigured, it will be configured automatically to be a power of 2 close to the number of configured threads in multi-threaded environments. Default: (unconfigured) 🧚
num-queries-per-thread: <number> — The number of queries that every thread will service simultaneously. If more queries arrive that need servicing, and no queries can be jostled out (see jostle-timeout), then the queries are dropped. This forces the client to resend after a timeout; allowing the server time to work on the existing queries. Default depends on compile options. Default: 2048 (libevent) / 512 (minievent) / 24 (windows) 💀
jostle-timeout: <msec> — Timeout used when the server is very busy. Set to a value that usually results in one roundtrip to the authority servers. If too many queries arrive, then 50% of the queries are allowed to run to completion, and the other 50% are replaced with the new incoming query if they have already spent more than their allowed time. This protects against denial of service by slow queries or high query rates. The effect is that the qps for long-lasting queries is about: (num-queries-per-thread / 2) / (average time for such long queries) qps. The qps for short queries can be about: (num-queries-per-thread / 2) / (jostle-timeout in whole seconds) qps per thread — about (2048/2)*5 = 5120 qps by default. Default: 200 🔥
delay-close: <msec> — Extra delay for timeouted UDP ports before they are closed, in msec. This prevents very delayed answer packets from the upstream (recursive) servers from bouncing against closed ports and setting off all sort of close-port counters, with eg. 1500 msec. When timeouts happen you need extra sockets, it checks the ID and remote IP of packets, and unwanted packets are added to the unwanted packet counter. Default: 0 (disabled) 💅
udp-connect: <yes or no> — Perform connect(2) for UDP sockets that mitigates ICMP side channel leakage. Default: yes ⭐
unknown-server-time-limit: <msec> — The wait time in msec for waiting for an unknown server to reply. Increase this if you are behind a slow satellite link, to eg. 1128. That would then avoid re-querying every initial query because it times out. Default: 376 🌈
discard-timeout: <msec> — The wait time in msec where recursion requests are dropped. This is to stop a large number of replies from accumulating. They receive no reply, the work item continues to recurse. For UDP the replies are dropped, for stream connections the reply is not dropped if the stream connection is still open ready to receive answers. It is nice to be a bit larger than serve-expired-client-timeout if that is enabled. A value of 1900 msec is suggested. The value 0 disables it. Default: 1900 🍓
wait-limit: <number> — The number of replies that can wait for recursion, for an IP address. This makes a ratelimit per IP address of waiting replies for recursion. It stops very large amounts of queries waiting to be returned to one destination. The value 0 disables all wait limits. Default: 1000 🫶
wait-limit-cookie: <number> — The number of replies that can wait for recursion, for an IP address that sent the query with a valid DNS Cookie. Since the cookie already validates the client address, this option allows to override a configured wait-limit value usually with a higher one for cookie validated queries. The value 0 disables wait limits for cookie validated queries. Default: 10000 💖
wait-limit-netblock: <netblock> <number> — The wait limit for the netblock. If not given the wait-limit value is used. The most specific netblock is used to determine the limit. Useful for overriding the default for a specific, group or individual, server. The value -1 disables wait limits for the netblock. By default the loopback has a wait limit netblock of -1, it is not limited, because it is separated from the rest of network for spoofed packets. The loopback addresses 127.0.0.0/8 and ::1/128 are default at -1. Default: (none) 🌸
wait-limit-cookie-netblock: <netblock> <number> — The wait limit for the netblock, when the query has a DNS Cookie. If not given, the wait-limit-cookie value is used. The value -1 disables wait limits for the netblock. The loopback addresses 127.0.0.0/8 and ::1/128 are default at -1. Default: (none) 🦋
so-rcvbuf: <number> — If not 0, then set the SO_RCVBUF socket option to get more buffer space on UDP port 53 incoming queries. So that short spikes on busy servers do not drop packets (see counter in netstat -su). Otherwise, the number of bytes to ask for, try "4m" on a busy server. The OS caps it at a maximum, on linux Unbound needs root permission to bypass the limit, or the admin can use sysctl net.core.rmem_max. On BSD change kern.ipc.maxsockbuf in /etc/sysctl.conf. On OpenBSD change header and recompile kernel. On Solaris ndd -set /dev/udp udp_max_buf 8388608. Default: 0 (use system value) 🔥
so-sndbuf: <number> — If not 0, then set the SO_SNDBUF socket option to get more buffer space on UDP port 53 outgoing responses. This for very busy servers handles spikes in answer traffic, otherwise: send: resource temporarily unavailable 💀

…and yeah, if it can get logged, the buffer overrun is also visible with netstat -su 💅. If set to 0, it uses the system value. You specify the number of bytes to ask for — try "8m" on a very busy server. It needs some space to be able to deal with packets that wait for local address resolution, from like ARP and NDP discovery, before they are sent out, hence it is elevated above the system default by default. The OS caps it at a maximum; on Linux, Unbound needs root permission to bypass the limit, or the admin can use sysctl net.core.wmem_max. On BSD and Solaris, changes are similar to so-rcvbuf. Default: 4m.

so-reuseport: <yes or no> — okay so if it’s yes, Unbound opens dedicated listening sockets for incoming queries for each thread and tries to set the SO_REUSEPORT socket option on each socket. May distribute incoming queries to threads more evenly. On Linux it is supported in kernels >= 3.9. On other systems, FreeBSD, OSX it may also work. You can enable it (on any platform and kernel); it then attempts to open the port and passes the option if it was available at compile time. If that works, it is used; if it fails, it continues silently (unless verbosity 3) without the option. At extreme load it could be better to turn it off to distribute the queries evenly, reported for Linux systems (4.4.x). Default: yes.
ip-transparent: <yes or no> — if yes, use the IP_TRANSPARENT socket option on sockets where Unbound is listening for incoming traffic. Allows you to bind to non-local interfaces — for example, for non-existent IP addresses that are going to exist later on, with host failover configuration. This is a lot like interface-automatic, but that one services all interfaces and with this option you can select which (future) interfaces Unbound provides service on. This option needs Unbound to be started with root permissions on some systems. The option uses IP_BINDANY on FreeBSD systems and SO_BINDANY on OpenBSD systems. Default: no.
ip-freebind: <yes or no> — if yes, use the IP_FREEBIND socket option on sockets where Unbound is listening to incoming traffic. Allows you to bind to IP addresses that are nonlocal or do not exist, like when the network interface or IP address is down. Exists only on Linux, where the similar ip-transparent option is also available. Default: no.
ip-dscp: <number> — the value of the Differentiated Services Codepoint (DSCP) in the differentiated services field (DS) of the outgoing IP packet headers. The field replaces the outdated IPv4 Type-Of-Service field and the IPv6 traffic class field.
rrset-cache-size: <number> — number of bytes size of the RRset cache. A plain number is in bytes; append 'k', 'm' or 'g' for kilobytes, megabytes or gigabytes (1024*1024 bytes in a megabyte). Default: 4m.
rrset-cache-slabs: <number> — number of slabs in the RRset cache. Slabs reduce lock contention by threads. Must be set to a power of 2. Setting (close) to the number of cpus is a fairly good setting. If left unconfigured, it will be configured automatically to be a power of 2 close to the number of configured threads in multi-threaded environments. Default: (unconfigured).
cache-max-ttl: <seconds> — time to live maximum for RRsets and messages in the cache. When the TTL expires, the cache item has expired. Can be set lower to force the resolver to query for data often, and not trust (very large) TTL values. Downstream clients also see the lower TTL. Default: 86400 (1 day).
cache-min-ttl: <seconds> — time to live minimum for RRsets and messages in the cache. If the minimum kicks in, the data is cached for longer than the domain owner intended, and thus less queries are made to look up the data. Zero makes sure the data in the cache is as the domain owner intended; higher values, especially more than an hour or so, can lead to trouble as the data in the cache does not match up with the actual data any more. Default: 0 (disabled).
cache-max-negative-ttl: <seconds> — time to live maximum for negative responses; these have a SOA in the authority section that is limited in time. This applies to NXDOMAIN and NODATA answers. Default: 3600.
cache-min-negative-ttl: <seconds> — time to live minimum for negative responses; these have a SOA in the authority section that is limited in time. If this is disabled and cache-min-ttl is configured, it will take effect instead. In that case you can set this to 1 to honor the upstream TTL. This applies to NXDOMAIN and NODATA answers. Default: 0 (disabled).
infra-host-ttl: <seconds> — time to live for entries in the host cache. The host cache contains roundtrip timing, lameness and EDNS support information. Default: 900.
infra-cache-slabs: <number> — number of slabs in the infrastructure cache. Slabs reduce lock contention by threads. Must be set to a power of 2. Setting (close) to the number of cpus is a fairly good setting. If left unconfigured, it will be configured automatically to be a power of 2 close to the number of configured threads in multi-threaded environments. Default: (unconfigured).
infra-cache-numhosts: <number> — number of hosts for which information is cached. Default: 10000.
infra-cache-min-rtt: <msec> — lower limit for dynamic retransmit timeout calculation in infrastructure cache. Increase this value if using forwarders needing more time to do recursive name resolution. Default: 50.
infra-cache-max-rtt: <msec> — upper limit for dynamic retransmit timeout calculation in infrastructure cache. Default: 120000 (2 minutes).
infra-keep-probing: <yes or no> — if enabled, the server keeps probing hosts that are down, in the one probe at a time regime. Hosts that are down, eg. they did not respond during the one probe at a time period, are marked as down and it may take infra-host-ttl time to get probed again. Default: no.
define-tag: "<list of tags>" — define the tags that can be used with local-zone and access-control. Enclose the list between quotes ("") and put spaces between tags.
do-ip4: <yes or no> — enable or disable whether IPv4 queries are answered or issued. Default: yes.
do-ip6: <yes or no> — enable or disable whether IPv6 queries are answered or issued. If disabled, queries are not answered on IPv6, and queries are not sent on IPv6 to the internet nameservers. With this option you can disable the IPv6 transport for sending DNS traffic; it does not impact the contents of the DNS traffic, which may have IPv4 (A) and IPv6 (AAAA) addresses in it. Default: yes.
prefer-ip4: <yes or no> — if enabled, prefer IPv4 transport for sending DNS queries to internet nameservers. Useful if the IPv6 netblock the server has, the entire /64 of that is not owned by one operator and the reputation of the netblock /64 is an issue; using IPv4 then uses the IPv4 filters that the upstream servers have. Default: no.
prefer-ip6: <yes or no> — if enabled, prefer IPv6 transport for sending DNS queries to internet nameservers. Default: no.
do-udp: <yes or no> — enable or disable whether UDP queries are answered or issued. Default: yes.
do-tcp: <yes or no> — enable or disable whether TCP queries are answered or issued. Default: yes.
tcp-mss: <number> — maximum segment size (MSS) of TCP socket on which the server responds to queries. Value lower than common MSS on Ethernet (1220 for example) will address path MTU problem. Note that not all platform supports socket option to set MSS (TCP_MAXSEG). Default is system default MSS determined by interface MTU and negotiation between server and client.
outgoing-tcp-mss: <number> — maximum segment size (MSS) of TCP socket for outgoing queries (from Unbound to other servers). Value lower than common MSS on Ethernet (1220 for example) will address path MTU problem. Note that not all platform supports socket option to set MSS (TCP_MAXSEG). Default is system default MSS determined by interface MTU and negotiation between Unbound and other servers.
tcp-idle-timeout: <msec> — the period Unbound will wait for a query on a TCP connection. If this timeout expires Unbound closes the connection. When the number of free incoming TCP buffers falls below 50% of the total number configured, the option value used is progressively reduced, first to 1% of the configured value, then to 0.2% of the configured value if the number of free buffers falls below 35% of the total number configured, and finally to 0 if the number of free buffers falls below 20% of the total number configured. A minimum timeout of 200 milliseconds is observed regardless of the option value used. It will be overridden by edns-tcp-keepalive-timeout if edns-tcp-keepalive is enabled. Default: 30000 (30 seconds).
tcp-reuse-timeout: <msec> — the period Unbound will keep TCP persistent connections open to authority servers. Default: 60000 (60 seconds).
max-reuse-tcp-queries: <number> — the maximum number of queries that can be sent on a persistent TCP connection. Default: 200.
tcp-auth-query-timeout: <number> — timeout in milliseconds for TCP queries to auth servers. Default: 3000 (3 seconds).
edns-tcp-keepalive: <yes or no> — enable or disable EDNS TCP Keepalive. Default: no.
edns-tcp-keepalive-timeout: <msec> — overrides tcp-idle-timeout when edns-tcp-keepalive is enabled. If the client supports the EDNS TCP Keepalive option, if the client supports the EDNS TCP Keepalive option 😭, Unbound sends the timeout value to the client to encourage it to close the connection before the server times out. Default: 120000 (2 minutes).
sock-queue-timeout: <sec> — okay so UDP queries can just be sitting in the socket buffer for ages, and those old ones can get dropped. You set the time in seconds, and lowkey 3 is a vibe-approved value for ignoring queries the client probably doesn't even want a reply to anymore 🙈 This happens when the host couldn't service queries for a while — like, Unbound wasn't running, and then gets enabled again. It uses timestamp socket options, and heads up: that socket option is only available on the Linux and FreeBSD platforms.

Default: 0 (disabled)

tcp-upstream: <yes or no> — turn on/off whether upstream queries use TCP only for transport. Super useful in tunneling scenarios ✨ If you set it to no, you can still specify TCP transport only for selected forward or stub zones using forward-tcp-upstream or stub-tcp-upstream respectively.

Default: no

udp-upstream-without-downstream: <yes or no> — enables UDP upstream even if do-udp: no is set. It's giving "TLS service provider" realness: you want no UDP downstream but you're still using UDP to fetch data upstream 💅

Default: no (no changes)

tls-upstream: <yes or no> — enable or disable whether the upstream queries use TLS only for transport. Useful in tunneling scenarios. The TLS contains plain DNS in TCP wireformat, and the other server must support this (see tls-service-key).

🚨 bestie listen: if you enable this, also configure a tls-cert-bundle or use tls-win-cert or tls-system-cert to load CA certs — otherwise the connections cannot be authenticated. no cap.

This option enables TLS for all of them, but if you do not set this you can configure TLS specifically for some forward zones with forward-tls-upstream. And also with stub-tls-upstream. If the tls-upstream option is enabled, it is for all the forwards and stubs, where the forward-tls-upstream and stub-tls-upstream options are ignored, as if they had been set to yes.

Default: no

ssl-upstream: <yes or no> — alternate syntax for tls-upstream. If both are present in the config file the last is used 👑
tls-service-key: <file> — if enabled, the server provides DNS-over-TLS or DNS-over-HTTPS service on the TCP ports marked implicitly or explicitly for these services with tls-port or https-port. The file must contain the private key for the TLS session; the public certificate is in the tls-service-pem file and it must also be specified if tls-service-key is specified 🔑

If the key is stored with root permissions or outside of chroot, then a change or enabling or disabling requires a restart (a reload is not enough 😭). But if the key file (and tls-service-pem file) are accessible, then they are read in on reload, and fast_reload. The server checks the modification time of the file (and the filename) to see if the file has changed for reload. The ports enabled implicitly or explicitly via tls-port and https-port do not provide normal DNS TCP service.

NOTE: Unbound needs to be compiled with libnghttp2 in order to provide DNS-over-HTTPS.

Default: "" (disabled)

ssl-service-key: <file> — alternate syntax for tls-service-key.
tls-service-pem: <file> — the public key certificate pem file for the tls service 🎀

Default: "" (disabled)

ssl-service-pem: <file> — alternate syntax for tls-service-pem.
tls-port: <number> — the port number on which to provide TCP TLS service. Only interfaces configured with that port number as @number get the TLS service.

Default: 853

ssl-port: <number> — alternate syntax for tls-port.
tls-cert-bundle: <file> — if null or "", no file is used. Set it to the certificate bundle file, for example /etc/pki/tls/certs/ca-bundle.crt. These certificates are used for authenticating connections made to outside peers — for example auth-zone urls, and also DNS-over-TLS connections. It is read at start up before permission drop and chroot 🌸

Default: "" (disabled)

ssl-cert-bundle: <file> — alternate syntax for tls-cert-bundle.
tls-win-cert: <yes or no> — add the system certificates to the cert bundle certificates for authentication. If no cert bundle, it uses only these certificates. On windows this option uses the certificates from the cert store. Use the tls-cert-bundle option on other systems. On other systems, this option enables the system certificates 🪟

Default: no

tls-system-cert: <yes or no> — this is the same as the tls-win-cert option, under a different name. Because it is not windows specific 💫
tls-additional-port: <portnr> — list port numbers as tls-additional-port, and when interfaces are defined, eg. with the @port suffix, as this port number, they provide DNS-over-TLS service. Can list multiple, each on a new statement 🔢
tls-session-ticket-keys: <file> — if not "", lists files with 80 bytes of random contents that are used to perform TLS session resumption for clients using the Unbound server. These files contain the secret key for the TLS session tickets. First key use to encrypt and decrypt TLS session tickets. Other keys use to decrypt only 🎟️

With this you can roll over to new keys, by generating a new first file and allowing decrypt of the old file by listing it after the first file for some time — after the wait, clients are not using the old key any more and the old key can be removed. One way to create the file is:

dd if=/dev/random bs=1 count=80 of=ticket.dat

The first 16 bytes should be different from the old one if you create a second key, that is the name used to identify the key. Then there is 32 bytes random data for an AES key and then 32 bytes random data for the HMAC key.

Default: ""

tls-ciphers: <string with cipher list> — set the list of ciphers to allow when serving TLS. Use "" for default ciphers.

Default: ""

tls-ciphersuites: <string with ciphersuites list> — set the list of ciphersuites to allow when serving TLS. This is for newer TLS 1.3 connections. Use "" for default ciphersuites 🌈

Default: ""

tls-use-sni: <yes or no> — enable or disable sending the SNI extension on TLS connections.

NOTE: Changing the value requires a restart.

Default: yes

tls-protocols: "<list of protocols>" — specify the allowed TLS protocol versions to use, in no particular order. Possible values are TLSv1.2 and TLSv1.3. Enclose list of protocols in quotes ("") and put spaces between them 🦋

NOTE: Changing the value requires a restart.

Default: "TLSv1.2 TLSv1.3"

pad-responses: <yes or no> — if enabled, TLS serviced queries that contained an EDNS Padding option will cause responses padded to the closest multiple of the size specified in pad-responses-block-size 🎈

Default: yes

pad-responses-block-size: <number> — the block size with which to pad responses serviced over TLS. Only responses to padded queries will be padded.

Default: 468

pad-queries: <yes or no> — if enabled, all queries sent over TLS upstreams will be padded to the closest multiple of the size specified in pad-queries-block-size.

Default: yes

pad-queries-block-size: <number> — the block size with which to pad queries sent over TLS upstreams.

Default: 128

https-port: <number> — the port number on which to provide DNS-over-HTTPS service. Only interfaces configured with that port number as @number get the HTTPS service 🔒

Default: 443

http-endpoint: <endpoint string> — the HTTP endpoint to provide DNS-over-HTTPS service on.

Default: /dns-query

http-max-streams: <number of streams> — number used in the SETTINGS_MAX_CONCURRENT_STREAMS parameter in the HTTP/2 SETTINGS frame for DNS-over-HTTPS connections.

Default: 100

http-query-buffer-size: <size in bytes> — maximum number of bytes used for all HTTP/2 query buffers combined. These buffers contain (partial) DNS queries waiting for request stream completion. An RST_STREAM frame will be send to streams exceeding this limit 💀 A plain number is in bytes, append 'k', 'm' or 'g' for kilobytes, megabytes or gigabytes (1024*1024 bytes in a megabyte).

Default: 4m

http-response-buffer-size: <size in bytes> — maximum number of bytes used for all HTTP/2 response buffers combined. These buffers contain DNS responses waiting to be written back to the clients. An RST_STREAM frame will be send to streams exceeding this limit. A plain number is in bytes, append 'k', 'm' or 'g' for kilobytes, megabytes or gigabytes (1024*1024 bytes in a megabyte).

Default: 4m

http-nodelay: <yes or no> — set TCP_NODELAY socket option on sockets used to provide DNS-over-HTTPS service. Ignored if the option is not available ⚡

Default: yes

http-notls-downstream: <yes or no> — disable use of TLS for the downstream DNS-over-HTTP connections. Useful for local back end servers.

Default: no

proxy-protocol-port: <portnr> — list port numbers as proxy-protocol-port, and when interfaces are defined, eg. with the @port suffix, as this port number, they support and expect PROXYv2 🫶

In this case the proxy address will only be used for the network communication and initial ACL (check if the proxy itself is denied/refused by configuration).

The proxied address (if any) will then be used as the true client address and will be used where applicable for logging, ACL, DNSTAP, RPZ and IP ratelimiting 👑

PROXYv2 is supported for UDP and TCP/TLS listening interfaces.

🚨 There is no support for PROXYv2 on a DoH, DoQ or DNSCrypt listening interface. periodt.

Can list multiple, each on a new statement.

quic-port: <number> — okay so this is the port where you serve DNS-over-QUIC 💅. Only interfaces configured with that port number as @number get the QUIC service. The interface uses QUIC for the UDP traffic on that port number. If it is set to 0, the server does not init QUIC code, and QUIC is disabled. This is similar to if QUIC is not in use, but then explicitly.

Default: 853

quic-size: <size in bytes> — max number of bytes for all QUIC buffers and data combined, bestie. A plain number is in bytes; append 'k', 'm' or 'g' for kilobytes, megabytes or gigabytes (1024*1024 bytes in a megabyte). New connections receive connection refused when the limit is exceeded. New streams are reset when the limit is exceeded.

Default: 8m

use-systemd: <yes or no> — enable or disable systemd socket activation. ✨

Default: no

do-daemonize: <yes or no> — enable or disable whether the Unbound server forks into the background as a daemon. Set the value to no when Unbound runs as systemd service.

Default: yes

tcp-connection-limit: <IP netblock> <limit> — allow up to limit simultaneous TCP connections from the given netblock. When at the limit, further connections are accepted but closed immediately. This option is experimental at this time 👀.

Default: (disabled)

access-control: <IP netblock> <action> — specify treatment of incoming queries from their originating IP address. Queries can be allowed to have access to this server that gives DNS answers, or refused, with other actions possible. The IP address range can be specified as a netblock; it is possible to give the statement several times in order to specify the treatment of different netblocks. The netblock is given as an IPv4 or IPv6 address with /size appended for a classless network block. The most specific netblock match is used; if none match refuse is used. The order of the access-control statements therefore does not matter. The action can be deny, refuse, allow, allow_setrd, allow_snoop, allow_cookie, deny_non_local or refuse_non_local. 🎀
- deny — stops queries from hosts from that netblock. 🚫
- refuse — stops queries too, but sends a DNS rcode REFUSED error message back. 😤
- allow — gives access to clients from that netblock. It gives only access for recursion clients (which is what almost all clients need). Non-recursive queries are refused.

The allow action does allow non-recursive queries to access the local-data that is configured. The reason is that this does not involve the Unbound server recursive lookup algorithm, and static data is served in the reply. This supports normal operations where non-recursive queries are made for the authoritative data. For non-recursive queries any replies from the dynamic cache are refused.

allow_setrd — ignores the recursion desired (RD) bit and treats all requests as if the recursion desired bit is set.

Note that this behavior violates RFC 1034 which states that a name server should never perform recursive service unless asked via the RD bit since this interferes with trouble shooting of name servers and their databases. This prohibited behavior may be useful if another DNS server must forward requests for specific zones to a resolver DNS server, but only supports stub domains and sends queries to the resolver DNS server with the RD bit cleared.

allow_snoop — gives non-recursive access too. This gives both recursive and non recursive access. The name allow_snoop refers to cache snooping, a technique to use non-recursive queries to examine the cache contents (for malicious acts). However, non-recursive queries can also be a valuable debugging tool (when you want to examine the cache contents).

In that case use allow_snoop for your administration host.

allow_cookie — allows access only to UDP queries that contain a valid DNS Cookie as specified in RFC 7873 and RFC 9018, when the answer-cookie option is enabled. UDP queries containing only a DNS Client Cookie and no Server Cookie, or an invalid DNS Cookie, will receive a BADCOOKIE response including a newly generated DNS Cookie, allowing clients to retry with that DNS Cookie. The allow_cookie action will also accept requests over stateful transports, regardless of the presence of an DNS Cookie and regardless of the answer-cookie setting. UDP queries without a DNS Cookie receive REFUSED responses with the TC flag set, that may trigger fall back to TCP for those clients. 🍪
deny_non_local — the deny_non_local action is for hosts that are only allowed to query for the authoritative local-data; they are not allowed full recursion but only the static data. Messages that are disallowed are dropped. 💀
refuse_non_local — the refuse_non_local action is for hosts that are only allowed to query for the authoritative local-data; they are not allowed full recursion but only the static data. Messages that are disallowed receive error code REFUSED.

By default only localhost (the 127.0.0.0/8 IP netblock, not the loopback interface) is implicitly allowed, the rest is refused. The default is refused, because that is protocol-friendly. The DNS protocol is not designed to handle dropped packets due to policy, and dropping may result in (possibly excessive) retried queries. 🫶

access-control-tag: <IP netblock> "<list of tags>" — assign tags to access-control elements. Clients using this access control element use localzones that are tagged with one of these tags.

Tags must be defined in define-tag. Enclose list of tags in quotes ("") and put spaces between tags.

If access-control-tag is configured for a netblock that does not have an access-control, an access-control element with action allow is configured for this netblock.

access-control-tag-action: <IP netblock> <tag> <action> — set action for particular tag for given access control element. If you have multiple tag values, the tag used to lookup the action is the first tag match between access-control-tag and local-zone-tag where "first" comes from the order of the define-tag values.
access-control-tag-data: <IP netblock> <tag> "<resource record string>" — set redirect data for particular tag for given access control element. ✨
access-control-view: <IP netblock> <view name> — set view for given access control element.
interface-action: <ip address or interface name [@port]> <action> — similar to access-control but for interfaces.

The action is the same as the ones defined under access-control.

Default action for interfaces is refuse. By default only localhost (the 127.0.0.0/8 IP netblock, not the loopback interface) is implicitly allowed through the default access-control behavior. This also means that any attempt to use the interface-*: options for the loopback interface will not work as they will be overridden by the implicit default "access-control: 127.0.0.0/8 allow" option.

NOTE: The interface needs to be already specified with interface and that any access-control*: option overrides all interface-*: options for targeted clients.

interface-tag: <ip address or interface name [@port]> <"list of tags"> — similar to access-control-tag but for interfaces.

NOTE: The interface needs to be already specified with interface and that any access-control*: option overrides all interface-*: options for targeted clients.

interface-tag-action: <ip address or interface name [@port]> <tag> <action> — similar to access-control-tag-action but for interfaces.

NOTE: The interface needs to be already specified with interface and that any access-control*: option overrides all interface-*: options for targeted clients.

interface-tag-data: <ip address or interface name [@port]> <tag> "<resource record string>" — similar to access-control-tag-data but for interfaces.

NOTE: The interface needs to be already specified with interface and that any access-control*: option overrides all interface-*: options for targeted clients.

interface-view: <ip address or interface name [@port]> <view name> — similar to access-control-view but for interfaces.

NOTE: The interface needs to be already specified with interface and that any access-control*: option overrides all interface-*: options for targeted clients.

chroot: <directory> — if chroot is enabled, you should pass the configfile (from the commandline) as a full path from the original root. After the chroot has been performed the now defunct portion of the config file path is removed to be able to reread the config after a reload. 🧚

All other file paths (working dir, logfile, roothints, and key files) can be specified in several ways: as an absolute path relative to the new root, as a relative path to the working directory, or as an absolute path relative to the original root. In the last case the path is adjusted to remove the unused portion.

The pidfile can be either a relative path to the working directory, or an absolute path relative to the original root. It is written just prior to chroot and dropping permissions. This allows the pidfile to be /var/run/unbound.pid and the chroot to be /var/unbound, for example. Note that Unbound is not able to remove the pidfile after termination when it is located outside of the chroot directory.

...and heads up, bestie, Unbound may also need to reach /dev/urandom from inside the chroot for entropy. 🌀

If you give it a value, a chroot(2) is done to that directory. If you give "", no chroot(2) happens at all.

Default: /etc/unbound

username: <name> — if you set this, the user privileges get dropped after the port is bound. Give username: "" and no user change happens, periodt.

Okay so here's the tea: if this user can't bind the port, reloads (triggered by signal HUP) will still keep the already-opened ports. But if you change the port number in the config file and that shiny new port needs privileges — the reload will straight-up fail, and you'll need a restart. 😭

Default: unbound

directory: <directory> — sets the working directory for the program. On Windows, the string "%EXECUTABLE%" tries to switch to the directory where unbound.exe lives. And if you put a server: directory: <directory> before any include file statements, those includes can be relative to the working directory. 🗂️

Default: /etc/unbound

logfile: <filename> — if "" is given, logging goes to stderr, or nowhere once it's daemonized (ghosted, basically 👻). The logfile is appended to, in this exact format:

[seconds since 1970] unbound[pid:tid]: type: message.

If this option is set, the use-syslog option gets internally flipped to no. The logfile is reopened (for append) when the config file is reread, on SIGHUP.

Default: "" (disabled)

use-syslog: <yes or no> — makes Unbound send log messages to the syslogd via syslog(3). The log facility LOG_DAEMON is used, with identity "unbound". When you set use-syslog: yes, the logfile setting gets overridden. ✨

Default: yes

log-identity: <string> — if "" is given, then the name of the executable, usually "unbound", is used in the log. Type a string to override it with that instead — super useful on systems running more than one Unbound instance with different configs, so you can actually tell the logs apart. 🔍

Default: ""

log-time-ascii: <yes or no> — sets logfile lines to use a timestamp in UTC ASCII. No effect if you're using syslog — in that case syslog handles the timestamp formatting in the log files itself.

Default: no (prints the seconds since 1970 in brackets)

log-time-iso: <yes or no> — logs time in ISO8601 format, but only if log-time-ascii: yes is also set. 🕰️

Default: no

log-queries: <yes or no> — prints one line per query to the log, with the log timestamp and IP address, name, type and class. Fair warning though: printing these lines takes time, which makes the server (significantly) slower. 😬 Odd (nonprintable) characters in names are printed as '?'.

Default: no

log-replies: <yes or no> — prints one line per reply to the log, with the log timestamp and IP address, name, type, class, return code, time to resolve, from cache and response size. Same deal: printing costs time and makes the server (significantly) slower. Odd (nonprintable) characters in names are printed as '?'.

Default: no

log-tag-queryreply: <yes or no> — prints the word 'query' and 'reply' alongside log-queries and log-replies. Makes filtering logs way easier, iykyk. 🏷️

Default: no (backwards compatible)

log-destaddr: <yes or no> — prints the destination address, port and type in the log-replies output. This disambiguates what type of traffic it was (eg. UDP or TCP) and which local port the traffic was sent to.

Default: no

log-local-actions: <yes or no> — prints log lines to inform you about local zone actions. These lines are like the local-zone type inform printouts, except they're also printed for the other types of local zones. 📍

Default: no

log-servfail: <yes or no> — prints log lines explaining why queries return SERVFAIL to clients. This is separate from the verbosity debug logs — much smaller, and printed at the error level, not the info level of debug info from verbosity.

Default: no

log-thread-id: <yes or no> — (only on Linux, and only when threads are available) logs the system-wide Linux thread ID instead of Unbound's internal thread counter. Chef's kiss when you're debugging with system tools. 🧵

Default: no

pidfile: <filename> — the process id gets written to the file. Default is /run/unbound.pid. So:

kill -HUP `cat /run/unbound.pid`

triggers a reload, and:

kill -TERM `cat /run/unbound.pid`

gracefully terminates. 💀

Default: /run/unbound.pid

root-hints: <filename> — reads the root hints from this file. Default is nothing, which means it uses builtin hints for the IN class. The file has the format of zone files, with root nameserver names and addresses only. Heads up: the default may become outdated when servers change — that's when a root hints file with specific servers comes in clutch.

Default: ""

hide-identity: <yes or no> — if enabled, 'id.server' and 'hostname.bind' queries are REFUSED. 🚫

Default: no

identity: <string> — sets the identity to report. If set to "", then the hostname of the server is returned.

Default: ""

hide-version: <yes or no> — if enabled, 'version.server' and 'version.bind' queries are REFUSED.

Default: no

version: <string> — sets the version to report. If set to "", then the package version is returned.

Default: ""

hide-http-user-agent: <yes or no> — if enabled, the HTTP header User-Agent is not set. ⚠️ Use with caution — some webserver configurations may reject HTTP requests lacking this header. If you need it, it's better to explicitly set the http-user-agent below instead.

Default: no

http-user-agent: <string> — sets the HTTP User-Agent header for outgoing HTTP requests. If set to "", then the package name and version are used.

Default: ""

nsid: <string> — adds the specified nsid to the EDNS section of the answer when queried with an NSID EDNS enabled packet. Goes in as a sequence of hex characters, or with an ascii_ prefix and then an ASCII string. 🌸

Default: (disabled)

hide-trustanchor: <yes or no> — if enabled, 'trustanchor.unbound' queries are REFUSED.

Default: no

target-fetch-policy: <"list of numbers"> — sets the target fetch policy Unbound uses to decide whether it should fetch nameserver target addresses opportunistically. The policy is described per dependency depth. 🎯

The number of values determines the maximum dependency depth that Unbound will pursue in answering a query. A value of -1 means fetch all targets opportunistically for that dependency depth. A value of 0 means fetch on demand only. A positive value fetches that many targets opportunistically.

Enclose the list between quotes ("") and put spaces between numbers. Setting all zeroes, "0 0 0 0 0", gives behaviour closer to that of BIND 9, while setting "-1 -1 -1 -1 -1" gives behaviour rumoured to be closer to that of BIND 8. 📜

Default: "3 2 1 0 0"

harden-short-bufsize: <yes or no> — very small EDNS buffer sizes from queries are ignored. 🛡️

Default: yes (per RFC 6891)

harden-large-queries: <yes or no> — very large queries are ignored. Default is no, since it's legal protocol-wise to send these, and they could be necessary for operation if TSIG or EDNS payload is very large.

Default: no

harden-glue: <yes or no> — will trust glue only if it's within the server's authority. 🔒

Default: yes

harden-unverified-glue: <yes or no> — will trust only in-zone glue. Will try to resolve all out-of-zone (unverified) glue. Will fall back to the original glue if it can't resolve.

Default: no

harden-dnssec-stripped: <yes or no> — requires DNSSEC data for trust-anchored zones; if that data is absent, the zone becomes bogus. If turned off, and no DNSSEC data is received (or the DNSKEY data fails to validate), then the zone is made insecure — this behaves like there's no trust anchor at all. You could turn this off if you're sometimes behind an intrusive firewall (of some sort) that removes DNSSEC data from packets, or if a zone changes from signed to unsigned to badly signed a lot. ⚠️ But if you turn it off, you run the risk of a downgrade attack that disables security for a zone.

Default: yes

harden-below-nxdomain: <yes or no> — from RFC 8020 (title: "NXDOMAIN: There Really Is Nothing Underneath" — iconic name ngl 📛), returns NXDOMAIN to queries for a name below another name that's already known to be NXDOMAIN. DNSSEC mandates NOERROR for empty nonterminals, hence this is possible. Very old software might return NXDOMAIN for empty nonterminals (which usually happen for reverse IP address lookups), and thus may be incompatible with this. To try to avoid that, only DNSSEC-secure NXDOMAINs are used, because the old software doesn't have DNSSEC.

NOTE: the NXDOMAIN must be secure — that means NSEC3 with optout is insufficient. 🚨

Default: yes

harden-referral-path: <yes or no> — hardens the referral path by performing additional queries for infrastructure data. Validates the replies if trust anchors are configured and the zones are signed. This enforces DNSSEC validation on nameserver NS sets and the nameserver addresses encountered on the referral path to the answer. Default is off, because it burdens the authority servers, it's not RFC standard, and it could lead to performance problems from the extra query load it generates. Experimental option, bestie — handle with care. If you enable it, consider adding more numbers after the target-fetch-policy to increase the max depth that gets checked to. 🔬

Default: no

harden-algo-downgrade — <yes or no>. This is the anti-downgrade guard for when multiple algorithms are advertised in the DS record, bestie. It works by first choosing only the strongest DS digest type as per RFC 4509 (Unbound treats the highest algorithm as the strongest) and then expecting signatures from all the advertised signing algorithms from the chosen DS(es) to be present. If no, it allows any one supported algorithm to validate the zone, even if other advertised algorithms are broken. RFC 6840 mandates that zone signers must produce zones signed with all advertised algorithms, but sometimes they do not. RFC 6840 also clarifies that this requirement is not for validators and validators should accept any single valid path. It should thus be explicitly noted that this option violates RFC 6840 for DNSSEC validation and should only be used to perform a signature completeness test to support troubleshooting.

WARNING: Using this option may break DNSSEC resolution with non RFC 6840 conforming signers and/or in multi-signer configurations that don't send all the advertised signatures.

Default: no

harden-unknown-additional — <yes or no>. Harden against unknown records in the authority section and additional section. If no, such records are copied from the upstream and presented to the client together with the answer. If yes, it could hamper future protocol developments that want to add records. Lowkey a tradeoff, no cap.

Default: no

use-caps-for-id — <yes or no>. Use 0x20-encoded random bits in the query to foil spoof attempts. This perturbs the lowercase and uppercase of query names sent to authority servers and checks if the reply still has the correct casing. This feature is an experimental implementation of draft dns-0x20. It’s giving “extra security sprinkles” ✨

Default: no

caps-exempt — <domain>. Exempt the domain so that it does not receive caps-for-id perturbed queries. For domains that do not support 0x20 and also fail with fallback because they keep sending different answers, like some load balancers. Can be given multiple times, for different domains. Very “add it to the list” energy.
caps-whitelist — <domain>. Alternate syntax for caps-exempt. Same vibe, different spelling 💅
qname-minimisation — <yes or no>. Send minimum amount of information to upstream servers to enhance privacy. Only send minimum required labels of the QNAME and set QTYPE to A when possible. Best effort approach; full QNAME and original QTYPE will be sent when upstream replies with a RCODE other than NOERROR, except when receiving NXDOMAIN from a DNSSEC signed zone. Privacy main character behavior, honestly 🫶

Default: yes

qname-minimisation-strict — <yes or no>. QNAME minimisation in strict mode. Do not fall-back to sending full QNAME to potentially broken nameservers. A lot of domains will not be resolvable when this option is enabled. Only use if you know what you are doing. This option only has effect when qname-minimisation is enabled.

Default: no

aggressive-nsec — <yes or no>. Aggressive NSEC uses the DNSSEC NSEC chain to synthesize NXDOMAIN and other denials, using information from previous NXDOMAINs answers. It helps to reduce the query rate towards targets that get a very high nonexistent name lookup rate. Iconic cache-efficiency behavior 🔥

Default: yes

private-address — <IP address or subnet>. Give IPv4 or IPv6 addresses or classless subnets. These are addresses on your private network, and are not allowed to be returned for public internet names. Any occurrence of such addresses are removed from DNS answers. Additionally, the DNSSEC validator may mark the answers bogus. This protects against so-called DNS Rebinding, where a user browser is turned into a network proxy, allowing remote access through the browser to other parts of your private network.

The option removes resource records of types A, AAAA, SVCB and HTTPS that match the filter. Inside the SVCB and HTTPS records, the svcparams of type ipv4hint and ipv6hint are checked for matches.

Some names can be allowed to contain your private addresses, by default all the local-data that you configured is allowed to, and you can specify additional names using private-domain. No private addresses are enabled by default.

We consider to enable this for the RFC 1918 private IP address space by default in later releases. That would enable private addresses for 10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16, 169.254.0.0/16, fd00::/8 and fe80::/10, since the RFC standards say these addresses should not be visible on the public internet.

Turning on 127.0.0.0/8 would hinder many spamblocklists as they use that. Adding ::ffff:0:0/96 stops IPv4-mapped IPv6 addresses from bypassing the filter. So specific, so important 💜

private-domain — <domain name>. Allow this domain, and all its subdomains to contain private addresses. Give multiple times to allow multiple domain names to contain private addresses.

Default: (none)

unwanted-reply-threshold — <number>. If set, a total number of unwanted replies is kept track of in every thread. When it reaches the threshold, a defensive action is taken and a warning is printed to the log. The defensive action is to clear the rrset and message caches, hopefully flushing away any poison. A value of 10 million is suggested.

It is useful to add 0.0.0.0/8 and '::' to the do-not-query-address list. Otherwise they may be answered, from localhost, and the different source makes an unwanted reply that unnecessarily ticks up. The do-not-query-localhost option includes them, the zero subnets, when it is enabled.

Default: 0 (disabled)

do-not-query-address — <IP address>. Do not query the given IP address. Can be IPv4 or IPv6. Append /num to indicate a classless delegation netblock, for example like 10.2.3.4/24 or 2001::11/64.

Default: (none)

do-not-query-localhost — <yes or no>. If yes, localhost is added to the do-not-query-address entries, both IPv6 ::1 and IPv4 127.0.0.1/8. If no, then localhost can be used to send queries to.

Default: yes

prefetch — <yes or no>. If yes, cache hits on message cache elements that are on their last 10 percent of their TTL value trigger a prefetch to keep the cache up to date. Turning it on gives about 10 percent more traffic and load on the machine, but popular items do not expire from the cache. Your fave records stay fresh, periodt 💅

Default: no

prefetch-key — <yes or no>. If yes, fetch the DNSKEYs earlier in the validation process, when a DS record is encountered. This lowers the latency of requests. It does use a little more CPU. Also if the cache is set to 0, it is no use.

Default: no

deny-any — <yes or no>. If yes, deny queries of type ANY with an empty response. If disabled, Unbound responds with a short list of resource records if some can be found in the cache and makes the upstream type ANY query if there are none. The option stops the DNSSEC validation from processing, possibly lengthy, ANY responses, when the option is enabled.

Default: no

rrset-roundrobin — <yes or no>. If yes, Unbound rotates RRSet order in response (the random number is taken from the query ID, for speed and thread safety). Little shuffle moment 🌈

Default: yes

minimal-responses — <yes or no>. If yes, Unbound does not insert authority/additional sections into response messages when those sections are not required. This reduces response size significantly, and may avoid TCP fallback for some responses which may cause a slight speedup. The default is yes, even though the DNS protocol RFCs mandate these sections, and the additional content could save roundtrips for clients that use the additional content. However these sections are hardly used by clients. Enabling prefetch can benefit clients that need the additional content by trying to keep that content fresh in the cache.

Default: yes

disable-dnssec-lame-check — <yes or no>. If yes, disables the DNSSEC lameness check in the iterator. This check sees if RRSIGs are present in the answer, when DNSSEC is expected, and retries another authority if RRSIGs are unexpectedly missing. The validator will insist in RRSIGs for DNSSEC signed domains regardless of this setting, if a trust anchor is loaded.

Default: no

module-config — "<module names>". Module configuration, a list of module names separated by spaces, surround the string with quotes (""). The modules can be respip, validator, or iterator (and possibly more, see below).

NOTE: The ordering of the modules is significant, the order decides the order of processing.

Setting this to just "iterator" will result in a non-validating server. Setting this to "validator iterator" will turn on DNSSEC validation.

NOTE: You must also set trust-anchors for validation to be useful.

Adding respip to the front will cause RPZ processing to be done on all queries.

Most modules that need to be listed here have to be listed at the beginning of the line.

The subnetcache module has to be listed just before the iterator.

The python module can be listed in different places, it then processes the output of the module it is just before.

The dynlib module can be listed pretty much anywhere, it is only a very thin wrapper that allows dynamic libraries to run in its place.

Default: "validator iterator"

trust-anchor-file — <filename>. File with trusted keys for validation. Both DS and DNSKEY entries can appear in the file. The format of the file is the standard DNS Zone file format.

Default: "" (no trust anchor file)

auto-trust-anchor-file: <filename> — a file holding the trust anchor for one zone, tracked with RFC 5011 probes 💅. The probes run several times per month, so like… the machine has to be online frequently, no cap. The initial file can be one with contents as described in trust-anchor-file. The file gets written to when the anchor updates, so the Unbound user needs write permission. And not just write permission to the file — also to the directory it's in (so it can create a temporary file, which is necessary to deal with filesystem full events), and it must also be inside the chroot if that's a thing you're doing.

Default: "" (no auto trust anchor file)

trust-anchor: "<Resource Record>" — a DS or DNSKEY RR for a key to use for validation ✨. You can give multiple entries to specify multiple trusted keys, in addition to trust-anchor-file. The resource record goes in the same format dig(1) or drill(1) prints them — same format as in the zone file. Has to be on a single line, with "" around it. A TTL can be specified for ease of cut and paste, but it's ignored (rude but ok). A class can be specified, but class IN is the default.

Default: (none)

trusted-keys-file: <filename> — file with trusted keys for validation 🎀. Wanna specify more than one file? Do it with several entries, one file per entry. Like trust-anchor-file but the file format is different — it's BIND-9 style format, the trusted-keys { name flag proto algo "key"; }; clauses are read. You can use wildcards with this statement, and the wildcard is expanded on start and on reload, slay.

Default: "" (no trusted keys file)

trust-anchor-signaling: <yes or no> — send RFC 8145 key tag query after trust anchor priming 🦋.

Default: yes

root-key-sentinel: <yes or no> — root key trust anchor sentinel 🌸.

Default: yes

domain-insecure: <domain name> — sets <domain name> to be insecure; the DNSSEC chain of trust is ignored towards the <domain name> 💀. So a trust anchor above the domain name can not make the domain secure with a DS record — such a DS record is then ignored. Can be given multiple times to specify multiple domains that are treated as if unsigned. And if you set trust anchors for the domain, they override this setting (and the domain is secured) 👑.

This can be useful if you want to make sure a trust anchor for external lookups does not affect an (unsigned) internal domain — because a DS record externally can create validation failures for that internal domain 🙈.

Default: (none)

val-override-date: <rrsig-style date spec>

WARNING: debugging feature! 😭

If enabled by giving a RRSIG style date, that date is used for verifying RRSIG inception and expiration dates, instead of the current date. Do not set this unless you're debugging signature inception and expiration, fr fr. The value -1 ignores the date altogether, useful for some special applications.

Default: 0 (disabled)

val-sig-skew-min: <seconds> — minimum number of seconds of clock skew to apply to validated signatures ⭐. A value of 10% of the signature lifetime (expiration − inception) is used, capped by this setting. Default is 3600 (1 hour), which allows for daylight savings differences. Lower this value for more strict checking of short lived signatures.

Default: 3600 (1 hour)

val-sig-skew-max: <seconds> — maximum number of seconds of clock skew to apply to validated signatures 🌈. A value of 10% of the signature lifetime (expiration − inception) is used, capped by this setting. Default is 86400 (24 hours), which allows for timezone setting problems in stable domains. Setting both min and max very low disables the clock skew allowances; setting both min and max very high makes the validator check the signature timestamps less strictly 👀.

Default: 86400 (24 hours)

val-max-restart: <number> — the maximum number the validator should restart validation with another authority in case of failed validation 🔥.

Default: 5

val-bogus-ttl: <seconds> — the time to live for bogus data 💀. This is data that has failed validation, due to invalid signatures or other checks. The TTL from that data cannot be trusted, so this value is used instead. The time interval prevents repeated revalidation of bogus data.

Default: 60

val-clean-additional: <yes or no> — instruct the validator to remove data from the additional section of secure messages that are not signed properly 🧚. Messages that are insecure, bogus, indeterminate or unchecked are not affected. Use this setting to protect the users that rely on this validator for authentication from potentially bad data in the additional section.

Default: yes

val-log-level: <number> — have the validator print validation failures to the log, regardless of the verbosity setting 📝.

At 1, for every user query that fails a line is printed to the logs — this way you can monitor what happens with validation. Use a diagnosis tool, such as dig or drill, to find out why validation is failing for these queries.

At 2, not only the query that failed is printed but also the reason why Unbound thought it was wrong and which server sent the faulty data 👑

Default: 0 (disabled)

val-permissive-mode: <yes or no> — instruct the validator to mark bogus messages as indeterminate 🙈. The security checks are performed, but if the result is bogus (failed security), the reply is not withheld from the client with SERVFAIL as usual — the client receives the bogus data. For messages that are found to be secure the AD bit is set in replies. Logging is also performed as for full validation.

Default: no

ignore-cd-flag: <yes or no> — instruct Unbound to ignore the CD flag from clients and refuse to return bogus answers to them 💅. So the CD (Checking Disabled) flag does not disable checking any more. This is useful if legacy (w2008) servers that set the CD flag but cannot validate DNSSEC themselves are the clients, and then Unbound provides them with DNSSEC protection.

Default: no

disable-edns-do: <yes or no> — disable the EDNS DO flag in upstream requests 💀. Heads up: it breaks DNSSEC validation for Unbound's clients. This results in the upstream name servers not including DNSSEC records in their replies and could be helpful for devices that cannot handle DNSSEC information. When the option is enabled, clients that set the DO flag receive no EDNS record in the response, to indicate the lack of support to them. If this option is enabled but Unbound is already configured for DNSSEC validation (i.e., the validator module is enabled; that's the default) this option is implicitly turned off with a warning, as to not break DNSSEC validation in Unbound — the config literally protects you from yourself, iconic.

Default: no

serve-expired: <yes or no> — if enabled, Unbound attempts to serve old responses from cache with a TTL of serve-expired-reply-ttl in the response 🍓. By default the expired answer will be used after a resolution attempt errored out or is taking more than serve-expired-client-timeout to resolve.

Default: no

serve-expired-ttl: <seconds> — limit serving of expired responses to the configured seconds after expiration ⏳. 0 disables the limit. This option only applies when serve-expired is enabled. A suggested value per RFC 8767 is between 86400 (1 day) and 259200 (3 days). The default is 86400.

Default: 86400

serve-expired-ttl-reset: <yes or no> — set the TTL of expired records to the serve-expired-ttl value after a failed attempt to retrieve the record from upstream 🔁. This makes sure that the expired records will be served as long as there are queries for it.

Default: no

serve-expired-reply-ttl: <seconds> — TTL value to use when replying with expired data 🎀. If serve-expired-client-timeout is also used then it is RECOMMENDED to use 30 as the value (RFC 8767).

This value is capped by the original TTL of the record. Which means records with a higher original TTL than this value will use this value for expired replies, and records with a lower original TTL than this value will use their original TTL for expired replies ✨.

Default: 30

serve-expired-client-timeout: <msec> — time in milliseconds before replying to the client with expired data ⌛. This essentially enables the serve-stale behavior as specified in RFC 8767, that first tries to resolve before immediately responding with expired data. Setting this to 0 will disable this behavior and instead serve the expired record immediately from the cache before attempting to refresh it via resolution.

Default: 1800

serve-original-ttl: <yes or no> — if enabled, Unbound will always return the original TTL as received from the upstream name server rather than the decrementing TTL as stored in the cache 🌟. This feature may be useful if Unbound serves as a front-end to a hidden authoritative name server.

Enabling this feature does not impact cache expiry, it only changes the TTL Unbound embeds in responses to queries.

NOTE: enabling this feature implicitly disables enforcement of the configured minimum and maximum TTL, as it is assumed users who enable this feature do not want Unbound to change the TTL obtained from an upstream server 🙈

NOTE: the values set using cache-min-ttl and cache-max-ttl are ignored 💀

Default: no

val-nsec3-keysize-iterations: <"list of values"> — list of keysize and iteration count values, separated by spaces, surrounded by quotes 🧚. This determines the maximum allowed NSEC3 iteration count before a message is simply marked insecure instead of performing the many hashing iterations. The list must be in ascending order and have at least one entry. If you set it to "1024 65535" there is no restriction to NSEC3 iteration values.

NOTE: this table has to stay short, bestie — like, a very long list could cause slower operation. Keep it cute and compact.

Default: "1024 150 2048 150 4096 150"

zonemd-permissive-mode: <yes or no> — turn this on and ZONEMD verification failures are only logged, they do not block the zone and only return servfail. It's giving "let me test if this even works" energy, or "I just wanna be notified of a problem without disrupting service" energy. Vibes-based observability, basically.

Default: no

add-holddown: <seconds> — tells the auto-trust-anchor-file probe mechanism for RFC 5011 autotrust updates to add new trust anchors only after they've been visible for this long 🕰️

Default: 2592000 (30 days as per the RFC)

del-holddown: <seconds> — same probe mechanism, but this one says: remove revoked trust anchors only after they've been chilling in the revoked list for this long.

Default: 2592000 (30 days as per the RFC)

keep-missing: <seconds> — instructs the auto-trust-anchor-file probe mechanism for RFC 5011 autotrust updates to remove missing trust anchors after they've been unseen for this long. This cleans up the state file if the target zone doesn't do trust anchor revocation, so it makes the auto probe mechanism work with zones that perform regular (non-5011) rollovers 🧹 The value 0 does not remove missing anchors, as per the RFC.

Default: 31622400 (366 days)

permit-small-holddown: <yes or no> — a debug option 🙈 that lets the autotrust 5011 rollover timers assume very small values.

Default: no

key-cache-size: <number> — the number of bytes size of the key cache. A plain number is in bytes, append 'k', 'm' or 'g' for kilobytes, megabytes or gigabytes (1024*1024 bytes in a megabyte).

Default: 4m

key-cache-slabs: <number> — number of slabs in the key cache. Slabs reduce lock contention by threads ✨ Must be set to a power of 2. Setting it (close) to the number of cpus is a fairly good setting. If you leave it unconfigured, it gets configured automatically to be a power of 2 close to the number of configured threads in multi-threaded environments.

Default: (unconfigured)

neg-cache-size: <number> — number of bytes size of the aggressive negative cache 😤 Same deal: plain number is bytes, append 'k', 'm' or 'g' for kilobytes, megabytes or gigabytes (1024*1024 bytes in a megabyte).

Default: 1m

unblock-lan-zones: <yes or no> — if enabled, then for private address space the reverse lookups are no longer filtered 🎉 This lets Unbound, when it's running as dns service on a host where it provides service for that host, put out all of the queries for the 'lan' upstream. When enabled, only localhost, 127.0.0.1 reverse and ::1 reverse zones are configured with default local zones. Disable this option when Unbound is running as a (DHCP-) DNS network resolver for a group of machines, where such lookups should be filtered (RFC compliance) — doing that also stops potential data leakage about the local network to the upstream DNS servers 🔒

Default: no

insecure-lan-zones: <yes or no> — if enabled, then reverse lookups in private address space are not validated. This is usually required whenever unblock-lan-zones is used 👀

Default: no

local-zone: <zone> <type> — configure a local zone, bestie. The type determines the answer to give if there's no match from local-data. The types are deny, refuse, static, transparent, redirect, nodefault, typetransparent, inform, inform_deny, inform_redirect, always_transparent, block_a, block_aaaa, block_a_wdata, block_aaaa_wdata, always_refuse, always_nxdomain, always_null, noview — and they're all explained below 👇 After that the default settings are listed. Use local-data to enter data into the local zone. Answers for local zones are authoritative DNS answers 👑 By default the zones are class IN.

If you need more complicated authoritative data — with referrals, wildcards, CNAME/DNAME support, or DNSSEC authoritative service — set up a stub-zone for it, as detailed in the stub zone section below. A stub-zone can be used to have unbound send queries to another server, an authoritative server, to fetch the information. With a forward-zone, unbound sends queries to a server that is a recursive server to fetch the information. With an auth-zone a zone can be loaded from file and used — it can be used like a local zone for users downstream, or the auth-zone information can be used to fetch information from when resolving, like it's an upstream server 🌊 The forward-zone and auth-zone options are described in their sections below. And if you want to perform filtering of the information that the users can fetch, the local-zone and local-data statements allow for that — but also the rpz functionality can be used, described in the RPZ section 💅

🚫 the local-zone types, explained

deny — do not send an answer, drop the query. If there's a match from local data, the query is answered.
refuse — send an error message reply, with rcode REFUSED. If there's a match from local data, the query is answered.
static — if there's a match from local data, the query is answered. Otherwise the query is answered with NODATA or NXDOMAIN. For a negative answer a SOA is included in the answer if present as local-data for the zone apex domain.
transparent — if there's a match from local-data, the query is answered. Otherwise if the query has a different name, the query is resolved normally. If the query is for a name given in local-data but no such type of data is given in localdata, then a NOERROR NODATA answer is returned. If no local-zone is given, local-data causes a transparent zone to be created by default ✨
typetransparent — if there's a match from local data, the query is answered. If the query is for a different name, or for the same name but for a different type, the query is resolved normally. So, similar to transparent but types that are not listed in local data are resolved normally — so if an A record is in the local data, that does not cause a NODATA reply for AAAA queries 🙌
redirect — the query is answered from the local data for the zone name. There may be no local data beneath the zone name. This answers queries for the zone, and all subdomains of the zone, with the local data for the zone. It can be used to redirect a domain to return a different address record to the end user, with:

local-zone: "example.com." redirect
local-data: "example.com. A 127.0.0.1"

queries for www.example.com and www.foo.example.com are redirected, so that users with web browsers cannot access sites with suffix example.com 💀

A CNAME record can also be provided via local-data:

local-zone: "example.com." redirect
local-data: "example.com. CNAME www.example.org."

In that case, the CNAME is resolved and the answer includes resolved target records as well. The CNAME record has to be with the zone name of the local-zone, and there can be one CNAME, not more. The CNAME record has to be at the zone apex of the redirect zone — then it's used for redirection. The resolution proceeds with upstream DNS resolution, and that does not include the lookup in local zones. So the record is not able to point in local zones, but it can point to upstream DNS answers ⬆️

CNAME resolution is supported only in type redirect local-zone, and in type inform_redirect local-zone.

As different from CNAME records that are used elsewhere, in the redirect type local-zone it's supported that in the target of the record a wildcard label gets expanded to the query name, with for example: example.com. CNAME *.foo.net. gets expanded to www.example.com. CNAME www.example.com.foo.net. 🌀

inform — the query is answered normally, same as transparent. The client IP address (@portnumber) is printed to the logfile. The log message is:

timestamp, unbound-pid, info: zonename inform IP@port queryname type class.

This option can be used for normal resolution, but machines looking up infected names are logged, eg. to run antivirus on them 🦠

inform_deny — the query is dropped, like deny, and logged, like inform. Ie. find infected machines without answering the queries.
inform_redirect — the query is redirected, like redirect, and logged, like inform. Ie. answer queries with fixed data and also log the machines that ask 📓
always_transparent — like transparent, but ignores local data and resolves normally.
block_a — like transparent, but ignores local data and resolves normally all query types excluding A. For A queries it unconditionally returns NODATA. Useful in cases when there's a need to explicitly force all apps to use IPv6 protocol and avoid any queries to IPv4 🌈
block_aaaa — like transparent or block_a, but ignores local data and resolves normally all query types excluding AAAA. For AAAA queries it unconditionally returns NODATA. Useful in cases when there's a need to explicitly force all apps to use IPv4 protocol and avoid any queries to IPv6 🌸
block_a_wdata — like block_a, but uses local data if present. If there's local data that is returned, and it acts like transparent. For A queries it returns NODATA.

block_aaaa_wdata — literally block_aaaa but it peeks at local data first 👀 if there's local data that comes back, it acts like transparent. For AAAA queries it gives you NODATA. periodt.
always_refuse — like refuse, except it ghosts your local data and just refuses the query. It also blocks queries of type DS for the zone name. yeah that can break the DNSSEC chain of trust 😭 but it's refused anyway so. The DS block is there to block the zone more completely, bestie.
always_nxdomain — like static, but ignores local data and returns NXDOMAIN for the query.
always_nodata — like static, but ignores local data and returns NODATA for the query.
always_deny — like deny, but ignores local data and just drops the query. gone. poof. 💨
always_null — always returns 0.0.0.0 or ::0 for every name in the zone. Like redirect with zero data for A and AAAA. Ignores local data in the zone. used for some block lists ✨
noview — breaks out of that view and moves towards the global local zones for the answer to the query. If view-first is no, it'll resolve normally. If view-first is enabled, it'll break, perform that step, and check the global answers. For when the view has view-specific overrides but some zone has to be answered from global local zone contents 🫶
nodefault — used to turn off default contents for AS112 zones. The other types also turn off default contents for the zone. nodefault has no other effect than turning off default contents for the given zone. Use nodefault if you use exactly that zone; if you want to use a subzone, use transparent instead.

The default zones are localhost, reverse 127.0.0.1 and ::1, the home.arpa, resolver.arpa, service.arpa, onion, test, invalid and the AS112 zones. The AS112 zones are reverse DNS zones for private use and reserved IP addresses where the servers on the internet literally cannot give correct answers. They're configured by default to give NXDOMAIN (no reverse information) answers.

The defaults can be turned off by specifying your own local-zone of that name, or by using the nodefault type. Below is the list of default zone contents 👇

🏠 `localhost`

The IPv4 and IPv6 localhost information is given. NS and SOA records are provided for completeness and to satisfy some DNS update tools. Default content:

local-zone: "localhost." redirect
local-data: "localhost. 10800 IN NS localhost."
local-data: "localhost. 10800 IN SOA localhost. nobody.invalid. 1 3600 1200 604800 10800"
local-data: "localhost. 10800 IN A 127.0.0.1"
local-data: "localhost. 10800 IN AAAA ::1"

🔁 reverse IPv4 loopback

Default content:

local-zone: "127.in-addr.arpa." static
local-data: "127.in-addr.arpa. 10800 IN NS localhost."
local-data: "127.in-addr.arpa. 10800 IN SOA localhost. nobody.invalid. 1 3600 1200 604800 10800"
local-data: "1.0.0.127.in-addr.arpa. 10800 IN PTR localhost."

🔁 reverse IPv6 loopback

Default content:

local-zone: "1.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.ip6.arpa." static
local-data: "1.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.ip6.arpa. 10800 IN NS localhost."
local-data: "1.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.ip6.arpa. 10800 IN SOA localhost. nobody.invalid. 1 3600 1200 604800 10800"
local-data: "1.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.ip6.arpa. 10800 IN PTR localhost."

🏡 `home.arpa` (RFC 8375)

Default content:

local-zone: "home.arpa." static
local-data: "home.arpa. 10800 IN NS localhost."
local-data: "home.arpa. 10800 IN SOA localhost. nobody.invalid. 1 3600 1200 604800 10800"

🔎 `resolver.arpa` (RFC 9462)

Default content:

local-zone: "resolver.arpa." static
local-data: "resolver.arpa. 10800 IN NS localhost."
local-data: "resolver.arpa. 10800 IN SOA localhost. nobody.invalid. 1 3600 1200 604800 10800"

🛠️ `service.arpa` (draft-ietf-dnssd-srp-25)

Default content:

local-zone: "service.arpa." static
local-data: "service.arpa. 10800 IN NS localhost."
local-data: "service.arpa. 10800 IN SOA localhost. nobody.invalid. 1 3600 1200 604800 10800"

🧅 `onion` (RFC 7686)

Default content:

local-zone: "onion." static
local-data: "onion. 10800 IN NS localhost."
local-data: "onion. 10800 IN SOA localhost. nobody.invalid. 1 3600 1200 604800 10800"

🧪 `test` (RFC 6761)

Default content:

local-zone: "test." static
local-data: "test. 10800 IN NS localhost."
local-data: "test. 10800 IN SOA localhost. nobody.invalid. 1 3600 1200 604800 10800"

🚫 `invalid` (RFC 6761)

Default content:

local-zone: "invalid." static
local-data: "invalid. 10800 IN NS localhost."
local-data: "invalid. 10800 IN SOA localhost. nobody.invalid. 1 3600 1200 604800 10800"

🏢 reverse local use zones (RFC 1918)

Reverse data for zones 10.in-addr.arpa, 16.172.in-addr.arpa to 31.172.in-addr.arpa, 168.192.in-addr.arpa. The local-zone is set static and as local-data SOA and NS records are provided.

🌐 special-use IPv4 Addresses (RFC 3330)

Reverse data for zones 0.in-addr.arpa (this), 254.169.in-addr.arpa (link-local), 2.0.192.in-addr.arpa (TEST NET 1), 100.51.198.in-addr.arpa (TEST NET 2), 113.0.203.in-addr.arpa (TEST NET 3), 255.255.255.255.in-addr.arpa (broadcast). And from 64.100.in-addr.arpa to 127.100.in-addr.arpa (Shared Address Space).

💜 reverse IPv6 unspecified (RFC 4291)

Reverse data for zone 0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.ip6.arpa.

🦋 reverse IPv6 Locally Assigned Local Addresses (RFC 4193)

Reverse data for zone D.F.ip6.arpa.

🔗 reverse IPv6 Link Local Addresses (RFC 4291)

Reverse data for zones 8.E.F.ip6.arpa to B.E.F.ip6.arpa.

🌸 reverse IPv6 Example Prefix

Reverse data for zone 8.B.D.0.1.0.0.2.ip6.arpa. This zone is used for tutorials and examples. You can remove the block on this zone with:

local-zone: 8.B.D.0.1.0.0.2.ip6.arpa. nodefault

You can also selectively unblock a part of the zone by making that part transparent with a local-zone statement. This also works with the other default zones 💅

✨ `local-data: "<resource record string>"` ✨

Configure local data, which is served in reply to queries for it. The query has to match exactly unless you configure the local-zone as redirect. If it's not matched exactly, the local-zone type determines further processing. If local-data is configured that isn't a subdomain of a local-zone, a transparent local-zone is configured. For record types such as TXT, use single quotes, like this:

local-data: 'example. TXT "text"'

NOTE: if you need more complicated authoritative data — with referrals, wildcards, CNAME/DNAME support, or DNSSEC authoritative service — set up a stub-zone for it as detailed in the stub zone section below 👑

🌈 `local-data-ptr: "IPaddr name"` 🌈

Configure local data shorthand for a PTR record with the reversed IPv4 or IPv6 address and the host name. For example "192.0.2.4 www.example.com". TTL can be inserted like this: "2001:db8::4 7200 www.example.com"

🎀 `local-zone-tag: <zone> <"list of tags">` 🎀

Assign tags to local zones. Tagged localzones will only be applied when the used access-control element has a matching tag. Tags must be defined in define-tag. Enclose the list of tags in quotes ("") and put spaces between tags. When there are multiple tags, it checks if the intersection of the list of tags for the query and local-zone-tag is non-empty.

💫 `local-zone-override: <zone> <IP netblock> <type>` 💫

Override the local zone type for queries from addresses matching netblock. Use this localzone type — regardless of the type configured for the local zone (both tagged and untagged) and regardless of the type configured using access-control-tag-action. it's giving override energy.

🔥 `response-ip: <IP-netblock> <action>` 🔥

This requires use of the respip module.

If the IP address in an AAAA or A RR in the answer section of a response matches the specified IP netblock, the specified action will apply. action has generally the same semantics as that for access-control-tag-action, but there are some exceptions 👀

Actions for response-ip are different from those for local-zone in that for the former there's no point to conditions like "the query matches it but there is no local data". Because of this difference, the semantics of response-ip actions are modified or simplified as follows: the static, refuse, transparent, typetransparent, and nodefault actions are invalid for response-ip. Using any of these will cause the configuration to be rejected as faulty 💀 The deny action is non-conditional, i.e. it always results in dropping the corresponding query. The resolution result before applying the deny action is still cached and can be used for other queries.

🫶 `response-ip-data: <IP-netblock> <"resource record string">` 🫶

This requires use of the respip module.

For the response-ip-data continuation: this specifies the action data for response-ip with action being to redirect as specified by <"resource record string">. It's giving very access-control-tag-data energy, but bestie, it must be of either AAAA, A or CNAME types. If the <IP-netblock> is an IPv6/IPv4 prefix, the record must be AAAA/A respectively, unless it is a CNAME (which can be used for both versions of IP netblocks). If it is CNAME there must not be more than one response-ip-data for the same <IP-netblock>. Also, CNAME and other types of records must not coexist for the same <IP-netblock>, following the normal rules for CNAME records. The textual domain name for the CNAME does not have to be explicitly terminated with a dot (.); the root name is assumed to be the origin for the name. No cap, those CNAME rules are not optional 💅

response-ip-tag: <IP-netblock> <"list of tags"> — this requires use of the respip module. It assigns tags to response <IP-netblock>. If the IP address in an AAAA or A RR in the answer section of a response matches the specified <IP-netblock>, the specified tags are assigned to the IP address. Then, if an access-control-tag is defined for the client and it includes one of the tags for the response IP, the corresponding access-control-tag-action will apply. Tag matching rule is the same as that for access-control-tag and local-zone. Unlike local-zone-tag, response-ip-tag can be defined for an <IP-netblock> even if no response-ip is defined for that netblock. If multiple response-ip-tag options are specified for the same <IP-netblock> in different statements, all but the first will be ignored. However, this will not be flagged as a configuration error, but the result is probably not what was intended. iykyk 😭

Actions specified in an access-control-tag-action that has a matching tag with response-ip-tag can be those that are "invalid" for response-ip listed above, since access-control-tag-action can be shared with local zones. For these actions, if they behave differently depending on whether local data exists or not in case of local zones, the behavior for response-ip-data will generally result in NOERROR/NODATA instead of NXDOMAIN, since the response-ip data are inherently type specific, and non-existence of data does not indicate anything about the existence or non-existence of the qname itself. For example, if the matching tag action is static but there is no data for the corresponding response-ip configuration, then the result will be NOERROR/NODATA. The only case where NXDOMAIN is returned is when an always_nxdomain action applies.

ratelimit: <number or 0> — enable ratelimiting of queries sent to nameserver for performing recursion. 0 disables the feature. This option is experimental at this time. highkey important, lowkey experimental

The ratelimit is in queries per second that are allowed. More queries are turned away with an error (SERVFAIL). Cached responses are not ratelimited by this setting.

This stops recursive floods, eg. random query names, but not spoofed reflection floods. The zone of the query is determined by examining the nameservers for it, the zone name is used to keep track of the rate. For example, 1000 may be a suitable value to stop the server from being overloaded with random names, and keeps unbound from sending traffic to the nameservers for those zones.

It is intended to count the number of queries towards the nameservers for the zone, and keep those queries limited. When there is a delegation that needs a lot of lookups, those are charged in the counters for the destination, the target name, of the NS records. Since that is where the nameserver lookup queries are sent to. That keeps the target, the victim domain, from having many queries. With the ratelimit-factor, some genuine queries that are also made to the target zone, can filter through, and then end up in cache, where the genuine answers have a chance to collect, keeping up service to some extent.

NOTE: Configured forwarders are excluded from ratelimiting.

Default: 0

ratelimit-size: <memory size> — give the size of the data structure in which the current ongoing rates are kept track in. In bytes or use m(mega), k(kilo), g(giga). The ratelimit structure is small, so this data structure likely does not need to be large. small but mighty vibes 🫶

Default: 4m

ratelimit-slabs: <number> — number of slabs in the ratelimit tracking data structure. Slabs reduce lock contention by threads. Must be set to a power of 2. Setting (close) to the number of cpus is a fairly good setting. If left unconfigured, it will be configured automatically to be a power of 2 close to the number of configured threads in multi-threaded environments.

Default: (unconfigured)

ratelimit-factor: <number> — set the amount of queries to rate limit when the limit is exceeded. If set to 0, all queries are dropped for domains where the limit is exceeded. If set to another value, 1 in that number is allowed through to complete. Default is 10, allowing 1/10 traffic to flow normally. This can make ordinary queries complete (if repeatedly queried for), and enter the cache, whilst also mitigating the traffic flow by the factor given. the way this lets a lil bit of traffic sneak through 🧚

Default: 10

ratelimit-backoff: <yes or no> — if enabled, the ratelimit is treated as a hard failure instead of the default maximum allowed constant rate. When the limit is reached, traffic is ratelimited and demand continues to be kept track of for a 2 second rate window. No traffic is allowed, except for ratelimit-factor, until demand decreases below the configured ratelimit for a 2 second rate window. Useful to set ratelimit to a suspicious rate to aggressively limit unusually high traffic. hard failure mode is not playing around 💀

Default: no

ratelimit-for-domain: <domain> <number qps or 0> — override the global ratelimit for an exact match domain name with the listed number. You can give this for any number of names. For example, for a top-level-domain you may want to have a higher limit than other names. A value of 0 will disable ratelimiting for that domain.
ratelimit-below-domain: <domain> <number qps or 0> — override the global ratelimit for a domain name that ends in this name. You can give this multiple times, it then describes different settings in different parts of the namespace. The closest matching suffix is used to determine the qps limit. The rate for the exact matching domain name is not changed, use ratelimit-for-domain to set that, you might want to use different settings for a top-level-domain and subdomains. A value of 0 will disable ratelimiting for domain names that end in this name. suffix matching, but make it fashion ✨
ip-ratelimit: <number or 0> — enable global ratelimiting of queries accepted per ip address. This option is experimental at this time. The ratelimit is in queries per second that are allowed. More queries are completely dropped and will not receive a reply, SERVFAIL or otherwise. IP ratelimiting happens before looking in the cache. This may be useful for mitigating amplification attacks. Clients with a valid DNS Cookie will bypass the ratelimit. If a ratelimit for such clients is still needed, ip-ratelimit-cookie can be used instead.

Default: 0 (disabled)

ip-ratelimit-cookie: <number or 0> — enable global ratelimiting of queries accepted per IP address with a valid DNS Cookie. This option is experimental at this time. The ratelimit is in queries per second that are allowed. More queries are completely dropped and will not receive a reply, SERVFAIL or otherwise. IP ratelimiting happens before looking in the cache. This option could be useful in combination with allow_cookie, in an attempt to mitigate other amplification attacks than UDP reflections (e.g., attacks targeting Unbound itself) which are already handled with DNS Cookies. If used, the value is suggested to be higher than ip-ratelimit e.g., tenfold. cookie-havers get their own special lane 🍓

Default: 0 (disabled)

ip-ratelimit-size: <memory size> — give the size of the data structure in which the current ongoing rates are kept track in. In bytes or use m(mega), k(kilo), g(giga). The IP ratelimit structure is small, so this data structure likely does not need to be large.

Default: 4m

ip-ratelimit-slabs: <number> — number of slabs in the ip ratelimit tracking data structure. Slabs reduce lock contention by threads. Must be set to a power of 2. Setting (close) to the number of cpus is a fairly good setting. If left unconfigured, it will be configured automatically to be a power of 2 close to the number of configured threads in multi-threaded environments.

Default: (unconfigured)

ip-ratelimit-factor: <number> — set the amount of queries to rate limit when the limit is exceeded. If set to 0, all queries are dropped for addresses where the limit is exceeded. If set to another value, 1 in that number is allowed through to complete. Default is 10, allowing 1/10 traffic to flow normally. This can make ordinary queries complete (if repeatedly queried for), and enter the cache, whilst also mitigating the traffic flow by the factor given. same factor energy, different IP lane 🌈

Default: 10

ip-ratelimit-backoff: <yes or no> — if enabled, the rate limit is treated as a hard failure instead of the default maximum allowed constant rate. When the limit is reached, traffic is ratelimited and demand continues to be kept track of for a 2 second rate window. No traffic is allowed, except for ip-ratelimit-factor, until demand decreases below the configured ratelimit for a 2 second rate window. Useful to set ip-ratelimit to a suspicious rate to aggressively limit unusually high traffic. backoff mode: activated, but make it polite 👑

Default: no

outbound-msg-retry: <number> — how many times Unbound retries, per upstream nameserver in a delegation, when it gets a throwaway response. 💀 A timeout (no response at all) still counts against the retry counter, just so you know. And if you're using a forward/stub zone? Then this is the retry count per nameserver in that zone.

Default: 5

max-sent-count: <number> — the hard cap on outgoing queries Unbound will fire off while resolving one name, so big mood NS sets don't just loop forever. Hit the limit and it's SERVFAIL city 😭. Fun fact: the counter resets on query restarts (like CNAMEs) and on referrals.

Default: 32

max-query-restarts: <number> — hard limit on how many times Unbound is allowed to restart a query when it bumps into a CNAME record. Reach it and boom, SERVFAIL. This applies to chained CNAME records — not the sporadic ones that just happen to show up during the query's resolution journey. And when a CNAME chain wraps up, the counter tracking this limit gets reset ✨. Change this value with caution bestie, because a bigger number means long CNAME chains get accepted, and Unbound has to verify (resolve) every single link individually. 🫶

Default: 11

iter-scrub-ns: <number> — caps how many NS records are allowed in an rrset of type NS, courtesy of the iterator scrubber. It's literally there to protect the resolver's internals from overly large NS sets. 👑

Default: 20

iter-scrub-cname: <number> — caps the number of CNAME and DNAME records in an answer, again from the iterator scrubber. Protects the resolver's internals from overly long indirection chains. It clips off the rest of the reply packet right at that point ✂️.

Default: 11

iter-scrub-rrsig: <number> — caps how many RRSIGs are allowed for an RRset, from the iterator scrubber. Guards against an overly large number of RRSIGs, and clips off the rest of the RRSIG list right there.

Default: 8

max-global-quota: <number> — the limit on upstream queries sent out for one incoming query plus all its subqueries from recursion. Important: it is not reset during resolution. Exceed it and the query fails and the whole lookup process stops. 🙈

Default: 200

iter-scrub-promiscuous: <yes or no> — should the iterator scrubber remove promiscuous NS from positive answers? It protects against poisonous contents that could otherwise affect names in the same zone as a spoofed packet. iykyk 💅

Default: yes

fast-server-permil: <number> — tells Unbound how many times out of 1000 to pick from the set of fastest servers. 0 turns the whole feature off. A value of 900 means picking from the fastest servers 90 percent of the time, and doing normal exploration of random servers the rest of the time. When prefetch is enabled (or serve-expired), those prefetches are not sped up — because nobody's waiting on them, so it's actually a great moment to do server exploration 😌. The fast-server-num option sets the size of that fastest-servers set.

Default: 0

fast-server-num: <number> — sets how many servers are used for fast server selection. You only use the fastest specified number of servers when fast-server-permil turns this on. 🏎️

Default: 3

answer-cookie: <yes or no> — if enabled, Unbound answers requests that contain DNS Cookies, exactly as specified in RFC 7873 and RFC 9018. 🍪

Default: no

cookie-secret: "<128 bit hex string>" — the server's secret for DNS Cookie generation. Genuinely useful to set explicitly for servers in an anycast deployment, because they need to share the secret to verify each other's Server Cookies. An example hex string would be "000102030405060708090a0b0c0d0e0f". ✨

NOTE: this option is ignored if a cookie-secret-file is present. In that case the secrets from that file are used in DNS Cookie calculations.

Default: 128 bits random secret generated at startup time

cookie-secret-file: <filename> — the file the secrets are read from for DNS Cookie calculations. When this file exists, the secrets in it are used and the secret from the cookie-secret option is ignored. Enable it by setting a filename, like "/usr/local/etc/unbound_cookiesecrets.txt". The content of this file must be manipulated with the add_cookie_secret, drop_cookie_secret and activate_cookie_secret commands to the unbound-control(8) tool. Go read that manpage for how to do a safe cookie secret rollover, bestie 🫶.

Default: "" (disabled)

edns-client-string: <IP netblock> <string> — includes an EDNS0 option holding the configured ASCII string in queries whose destination address matches the configured <IP netblock>. You can use this option multiple times, and the most specific match wins. 🎯
edns-client-string-opcode: <opcode> — the EDNS0 option code for the edns-client-string option, from 0 to 65535. You should use a value from the 'Reserved for Local/Experimental' range (65001-65534). 🧚

Default: 65001

ede: <yes or no> — if enabled, Unbound responds with Extended DNS Error codes (RFC 8914). These EDEs give extra info alongside a response, mainly for DNS and DNSSEC errors — though not limited to those. 🌈

And when the val-log-level option is also set to 2, responses with Extended DNS Errors about DNSSEC failures will also carry a descriptive text message explaining why it failed. So helpful fr.

Default: no

ede-serve-expired: <yes or no> — if enabled, Unbound attaches an Extended DNS Error (RFC 8914) Code 3 - Stale Answer as an EDNS0 option to the expired response. 🍓

NOTE: ede: yes needs to be set as well for this to work.

Default: no

dns-error-reporting: <yes or no> — if enabled, Unbound sends DNS Error Reports (RFC 9567). The nameservers have to express support by attaching the Report-Channel EDNS0 option on their replies, specifying the reporting agent for the zone. Any errors hit during resolution that would make Unbound generate an Extended DNS Error (RFC 8914) get reported to that zone's reporting agent. 📮

The ede option does not need to be enabled for this to work.

It's advised that the qname-minimisation option is also enabled, to increase privacy on the outgoing reports. 💜

Default: no

🎛️ REMOTE CONTROL OPTIONS 🎛️

Okay so these options are part of the remote-control: section, and they're the declarations for the remote control facility. 💅

If this is enabled, the unbound-control(8) utility can send commands to the running Unbound server. The server uses these options to set up TLS security for the connection. The unbound-control(8) utility also reads this remote-control: section for options. And to set up the correct self-signed certificates, you use the unbound-control-setup(8) utility. ✨

control-enable: <yes or no> — this one enables remote control. If it's turned off, the server doesn't listen for control commands at all. 🚫

Default: no

control-interface: <IP address or interface name[@port] or path> — give IPv4 or IPv6 addresses, or a local socket path, to listen on for control commands. If you use an interface name instead of an IP address, the list of IP addresses on that interface is used. 🦋

By default localhost (127.0.0.1 and ::1) is what's listened to. Use 0.0.0.0 and ::0 to listen to all interfaces. If you change this and permissions have been dropped, you must restart the server for the change to take effect. 😤

If you set it to an absolute path, a unix domain socket is used. This socket doesn't use the certificates and keys, so those files need not be present 🎀. To restrict access, Unbound sets permissions on the file to the user and group that's configured, and the access bits are set so group members can access the control socket file. So put the users that need to access the socket in that group. Want even tighter access? Create a directory, put the control socket in it, and restrict access to that directory. 🔒

control-port: <port number> — the port number to listen on for IPv4 or IPv6 control interfaces.

NOTE: if you change this and permissions have been dropped, you must restart the server for the change to take effect.

Default: 8953

control-use-cert: <yes or no> — for a localhost control-interface you can disable the use of TLS by setting this option to "no". For local sockets, TLS is disabled and the value of this option is ignored. 💀

Default: yes

server-key-file: <private key file> — path to the server private key. This file is generated by the unbound-control-setup(8) utility. It's used by the Unbound server, but not by unbound-control(8). 🔑

Default: unbound_server.key

server-cert-file: <certificate file.pem> — path to the server self signed certificate. Generated by the unbound-control-setup(8) utility. This one is used by the Unbound server and also by unbound-control(8). 📜

Default: unbound_server.pem

control-key-file: <private key file> — path to the control client private key. Generated by the unbound-control-setup(8) utility, and used by unbound-control(8). 🗝️

Default: unbound_control.key

control-cert-file: <certificate file.pem> — path to the control client certificate. This certificate has to be signed with the server certificate. Generated by the unbound-control-setup(8) utility, and used by unbound-control(8). 👑

Default: unbound_control.pem

🦋 STUB ZONE OPTIONS 🦋

Okay so these are the options that live inside the stub-zone: section, bestie. Buckle up.

You can have, like, multiple stub-zone: sections — no cap. Each one has a name and zero or more hostnames or IP addresses. For the stub zone, this list of nameservers is what gets used. Class IN is assumed (always IN, we don't do drama here). And the servers should be authority servers, not recursors — for stub zones Unbound does the recursive processing itself, iykyk. 💅

So what's the whole point of a stub zone? It lets you configure authoritative data that the resolver can use but that can't be reached through the public internet servers. Super useful for company-local data or private zones. You set up an authoritative server on a different host (or a different port), and then you put a config entry in Unbound like this:

stub-addr: <ip address of host[@port]>

And then the Unbound resolver can access that data, without ever referring to the public internet for it. Slay. ✨

This setup also lets DNSSEC signed zones be served by that authoritative server — and in that case you can put a trusted key entry with the public key in the config, so Unbound can validate the data and set the AD bit on replies for the private zone (because authoritative servers do not set the AD bit, rude of them honestly). This setup makes Unbound capable of answering queries for the private zone, and it can even set the AD bit ('authentic') — but heads up, the AA ('authoritative') bit is not set on these replies. 🙈

Consider adding server statements for domain-insecure and for local-zone: <name> nodefault for the zone if it's a locally served zone. Why? The domain-insecure option stops DNSSEC from invalidating the zone. And the local-zone: nodefault (or transparent) option makes the (reverse-) zone bypass Unbound's filtering of RFC 1918 zones. Little details, big impact. 🎀

name: <domain name> — Name of the stub zone. This is the full domain name of the zone. 💖
stub-host: <domain name> — Name of the stub zone nameserver. It is itself resolved before it is used.

⚠️ CAUTION: If the domain (or a subdomain) from this zone is used as the host, it will unavoidably introduce a circular dependency on retrieving the IP addresses of the name server. In that case, it is suggested to use stub-addr instead. Alternatively, stub-first: yes can also work around the circular dependency by trying resolution outside of this zone. However this has the caveat that it would allow escaping this zone when any resolution attempt fails within this zone. 😭

To use a non-default port for DNS communication, append '@' with the port number. 🔥

If TLS is enabled, then you can append a '#' and a name, and then it'll check the TLS authentication certificates with that name.

If you combine the '@' and '#', the '@' comes first. If only '#' is used, the default port is the configured tls-port. 🧚

stub-addr: <IP address> — IP address of stub zone nameserver. Can be IPv4 or IPv6.

To use a non-default port for DNS communication, append '@' with the port number.

If TLS is enabled, then you can append a '#' and a name, and then it'll check the tls authentication certificates with that name.

If you combine the '@' and '#', the '@' comes first. If only '#' is used, the default port is the configured tls-port. 🌸

stub-prime: <yes or no> — If enabled, it performs NS set priming, which is similar to root hints, where it starts using the list of nameservers currently published by the zone. So if the hint list is slightly outdated, the resolver picks up a correct list online. Iconic behaviour honestly. 👑

Default: no

stub-first: <yes or no> — If enabled, a query is attempted without this stub section if it fails. Like, the data could not be retrieved and would have caused SERVFAIL because the servers are unreachable — instead it's tried without this stub section. This can lead to using less specific configured forward/stub/auth zones if any, or end up to otherwise normal recursive resolution for that particular query.

Default: no

stub-tls-upstream: <yes or no> — Enabled or disable whether the queries to this stub use TLS for transport.

Default: no

stub-ssl-upstream: <yes or no> — Alternate syntax for stub-tls-upstream.
stub-tcp-upstream: <yes or no> — If it is set to "yes" then upstream queries use TCP only for transport, regardless of global flag tcp-upstream.

Default: no

stub-no-cache: <yes or no> — If enabled, data inside the stub is not cached. This is useful when you want immediate changes to be visible. Fr fr, no stale vibes allowed. 💀

Default: no

💜 FORWARD ZONE OPTIONS 💜

These ones are part of the forward-zone: section, and honestly they're giving main character energy.

There may be multiple forward-zone: sections. Each with a name and zero or more hostnames or IP addresses. For the forward zone, this list of nameservers is used to forward the queries to. The servers listed as forward-host and forward-addr have to handle further recursion for the query. So those servers are not authority servers — they're (just like Unbound is) recursive servers too. Unbound does not perform recursion itself for the forward zone, it lets the remote server do it. Delegation queen. 👑

Class IN is assumed. CNAMEs are chased by Unbound itself, asking the remote server for every name in the indirection chain, to protect the local cache from illegal indirect referenced items. And a forward-zone entry with name "." and a forward-addr target will forward all queries to that other server (unless it can answer from the cache). 🌈

name: <domain name> — Name of the forward zone. This is the full domain name of the zone. ⭐
forward-host: <domain name> — Name of server to forward to. It is itself resolved before it is used.

⚠️ CAUTION: If the domain (or a subdomain) from this zone is used as the host, it will unavoidably introduce a circular dependency on retrieving the IP addresses of the name server. In that case, it is suggested to use forward-addr instead. Alternatively, forward-first: yes can also work around the circular dependency by trying resolution outside of this zone. However this has the caveat that it would allow escaping this zone when any resolution attempt fails within this zone. 😭

To use a non-default port for DNS communication, append '@' with the port number.

If TLS is enabled, then you can append a '#' and a name, and then it'll check the TLS authentication certificates with that name.

If you combine the '@' and '#', the '@' comes first. If only '#' is used, the default port is the configured tls-port. 🫶

forward-addr: <IP address> — IP address of server to forward to. Can be IPv4 or IPv6.

To use a non-default port for DNS communication, append '@' with the port number.

If TLS is enabled, then you can append a '#' and a name, and then it'll check the TLS authentication certificates with that name.

If you combine the '@' and '#', the '@' comes first. If only '#' is used, the default port is the configured tls-port.

At high verbosity it logs the TLS certificate, with TLS enabled. If you leave out the '#' and auth name from the forward-addr, any name is accepted. The cert must also match a CA from the tls-cert-bundle. Don't skip that part bestie, it matters. 🔥

forward-first: <yes or no> — If a forwarded query is met with a SERVFAIL error and this option is enabled, Unbound will fall back to less specific resolution. This can lead to using less specific configured forward/stub/auth zones if any, or end up to otherwise normal recursive resolution for that particular query.

Default: no

forward-tls-upstream: <yes or no> — Enabled or disable whether the queries to this forwarder use TLS for transport. And listen — if you enable this, also configure a tls-cert-bundle or use tls-win-cert to load CA certs, otherwise the connections cannot be authenticated. Like, the vibes will be insecure. 😬

Default: no

forward-ssl-upstream: <yes or no> — Alternate syntax for forward-tls-upstream.
forward-tcp-upstream: <yes or no> — If it is set to "yes" then upstream queries use TCP only for transport, regardless of global flag tcp-upstream.

Default: no

forward-no-cache: <yes or no> — If enabled, data inside the forward is not cached. This is useful when you want immediate changes to be visible. Fresh data only, periodt. ✨

Default: no

✨ AUTHORITY ZONE OPTIONS ✨

okay bestie, these options are part of the auth-zone: section. 🫶

So auth-zone: sections are how you set up authority zones, and every single one needs a name. You can have multiple ones by listing multiple auth-zone section clauses, each with a different name, covering that part of the namespace. When Unbound looks something up, it uses the authority zone whose name is closest to the name being looked up. Authority zones can be processed at two distinct, non-exclusive, configurable stages. ✨

With for-downstream: yes (the default), authority zones are processed after local-zones and before cache. When it's used like this, Unbound responds like an authority server with no further processing other than returning an answer from the zone contents. A notable example here is CNAME records, which are returned verbatim to downstream clients without further resolution. 💅

With for-upstream: yes (also default), authority zones are processed after the cache lookup, just before going to the network to fetch information for recursion. Used this way, they give you a local copy of an authority server that speeds up lookups for that data during resolving. 🦋

If both options are enabled (default), client queries for an authority zone are answered authoritatively from Unbound, while internal queries that need data from the authority zone consult the local zone data instead of going to the network. 🔥

An interesting configuration is for-downstream: no, for-upstream: yes — that allows for hyperlocal behavior where both client and internal queries consult the local zone data while resolving. In this case, the aforementioned CNAME example will result in a thoroughly resolved answer. 🎀

Authority zones can be read from a zonefile. And they can be kept updated via AXFR and IXFR. After an update the zonefile is rewritten. The update mechanism uses the SOA timer values and performs SOA UDP queries to detect zone changes. ⭐

If the update fetch fails, the timers in the SOA record are used to time another fetch attempt. Until the SOA expiry timer is reached. Then the zone is expired. When a zone is expired, queries are SERVFAIL, and any new serial number is accepted from the primary (even if older), and if fallback is enabled, the fallback activates to fetch from the upstream instead of the SERVFAIL. 💀

name: <zone name> — name of the authority zone.
primary: <IP address or host name> — where to download a copy of the zone from, with AXFR and IXFR. You can specify multiple primaries, and they are all tried if one fails. To use a non-default port for DNS communication, append '@' with the port number. You can append a '#' and a name, then AXFR over TLS can be used and the TLS authentication certificates will be checked with that name. If you combine the '@' and '#', the '@' comes first. If you point it at another Unbound instance, it would not work because that does not support AXFR/IXFR for the zone, but if you used url to download the zonefile as a text file from a webserver that would work.

CAUTION: If you specify the hostname, you cannot use the domain from the zonefile, because it may not have that when retrieving that data, instead use a plain IP address to avoid a circular dependency on retrieving that IP address.

master: <IP address or host name> — alternate syntax for primary.
url: <URL to zone file> — where to download a zonefile for the zone. With HTTP or HTTPS. An example for the url is: http://www.example.com/example.org.zone. Multiple url statements can be given, they are tried in turn. If only urls are given the SOA refresh timer is used to wait for making new downloads. If also primaries are listed, the primaries are first probed with UDP SOA queries to see if the SOA serial number has changed, reducing the number of downloads. If none of the urls work, the primaries are tried with IXFR and AXFR. For HTTPS, the tls-cert-bundle and the hostname from the url are used to authenticate the connection. If you specify a hostname in the URL, you cannot use the domain from the zonefile, because it may not have that when retrieving that data, instead use a plain IP address to avoid a circular dependency on retrieving that IP address. Avoid dependencies on name lookups by using a notation like "http://192.0.2.1/unbound-primaries/example.com.zone", with an explicit IP address.
allow-notify: <IP address or host name or netblockIP/prefix> — with allow-notify you can specify additional sources of notifies. When notified, the server attempts to first probe and then zone transfer. If the notify is from a primary, it first attempts that primary. Otherwise other primaries are attempted. If there are no primaries, but only urls, the file is downloaded when notified.

NOTE: The primaries from primary and url statements are allowed notify by default.

fallback-enabled: <yes or no> — if enabled, Unbound falls back to querying the internet as a resolver for this zone when lookups fail. For example for DNSSEC validation failures. Default: no.
for-downstream: <yes or no> — if enabled, Unbound serves authority responses to downstream clients for this zone. This option makes Unbound behave, for the queries with names in this zone, like one of the authority servers for that zone. Turn it off if you want Unbound to provide recursion for the zone but have a local copy of zone data. If for-downstream: no and for-upstream: yes are set, then Unbound will DNSSEC validate the contents of the zone before serving the zone contents to clients and store validation results in the cache. Default: yes.
for-upstream: <yes or no> — if enabled, Unbound fetches data from this data collection for answering recursion queries. Instead of sending queries over the internet to the authority servers for this zone, it'll fetch the data directly from the zone data. Turn it on when you want Unbound to provide recursion for downstream clients, and use the zone data as a local copy to speed up lookups. Default: yes.
zonemd-check: <yes or no> — enable this option to check ZONEMD records in the zone. The ZONEMD record is a checksum over the zone data. This includes glue in the zone and data from the zone file, and excludes comments from the zone file. When there is a DNSSEC chain of trust, DNSSEC signatures are checked too. Default: no.
zonemd-reject-absence: <yes or no> — enable this option to reject the absence of the ZONEMD record. Without it, when ZONEMD is not there it is not checked. It is useful to enable for a non-DNSSEC signed zone where the operator wants to require the verification of a ZONEMD, hence a missing ZONEMD is a failure. The action upon failure is controlled by the zonemd-permissive-mode option, for log only or also block the zone. Without the option, absence of a ZONEMD is only a failure when the zone is DNSSEC signed, and we have a trust anchor, and the DNSSEC verification of the absence of the ZONEMD fails. With the option enabled, the absence of a ZONEMD is always a failure, also for nonDNSSEC signed zones. Default: no.
zonefile: <filename> — the filename where the zone is stored. If not given then no zonefile is used. If the file does not exist or is empty, Unbound will attempt to fetch zone data (eg. from the primary servers).
max-transfer-size: <number> — number of bytes size of the maximum zone transfer size. Larger transfers, over AXFR, IXFR and HTTP, are not allowed. A plain number is in bytes, append 'k', 'm' or 'g' for kilobytes, megabytes or gigabytes (1024*1024 bytes in a megabyte). The value 0 disables the feature. Only consider for untrusted/misbehaving primaries that could hog resources and bring down the resolver. Default: 0.
max-transfer-time: <msec> — maximum time in milliseconds that a zone transfer is allowed to take from the start. The value 0 disables the feature. Only consider for untrusted/misbehaving primaries that could hog resources and bring down the resolver. Default: 0.

🌈 VIEW OPTIONS 🌈

These options are part of the view: section. 👑

There may be multiple view: sections. Each with a name and zero or more local-zone and local-data options. Views can also contain view-first, response-ip, response-ip-data and local-data-ptr options. A view can be mapped to requests by specifying the view name in an access-control-view option. Options from matching views will override global options. Global options will be used if no matching view is found, or when the matching view does not have the option specified. 💖

name: <view name> — name of the view. Must be unique. This name is used in the access-control-view option.
local-zone: <zone> <type> — view specific local zone elements. Has the same types and behaviour as the global local-zone elements. When there is at least one local-zone: specified and view-first: no is set, the default local-zones will be added to this view. Defaults can be disabled using the nodefault type. When view-first: yes is set or when a view does not have a local-zone, the global local-zone will be used including it's default zones.
local-data: "<resource record string>" — view specific local data elements. Has the same behaviour as the global local-data elements.
local-data-ptr: "IPaddr name" — view specific local-data-ptr elements. Has the same behaviour as the global local-data-ptr elements.
response-ip: <IP-netblock> <action> — this requires use of the respip module. Similar to response-ip but only applies to this view.
response-ip-data: <IP-netblock> <"resource record string"> — this requires use of the respip module. Similar to response-ip-data but only applies to this view.
view-first: <yes or no> — if enabled, it attempts to use the global local-zone and local-data if there is no match in the view specific options. Default: no.

💖 PYTHON MODULE OPTIONS 💖

These options are part of the python: section.

Okay so the python: section gives the settings for the python(1) script module. This module acts like the iterator and validator modules do, on queries and answers. To enable the script module it has to be compiled into the daemon, and the word python has to be put in the module-config option (usually first, or between the validator and iterator). Multiple instances of the python module are supported by adding the word python more than once.

If the chroot option is enabled, you should make sure Python's library directory structure is bind mounted in the new root environment, see mount(8). Also the python-script path should be specified as an absolute path relative to the new root, or as a relative path to the working directory.

python-script: <python file> — the script file to load. Repeat this option for every python module instance added to the module-config option. 🐍✨

🎀 DYNAMIC LIBRARY MODULE OPTIONS 🎀

These options are part of the dynlib: section.

The dynlib: section gives the settings for the dynlib module. This module is only a very small wrapper that allows dynamic modules to be loaded on runtime instead of being compiled into the application. To enable the dynlib module it has to be compiled into the daemon, and the word dynlib has to be put in the module-config option. Multiple instances of dynamic libraries are supported by adding the word dynlib more than once.

The dynlib-file path should be specified as an absolute path relative to the new path set by chroot, or as a relative path to the working directory.

dynlib-file: <dynlib file> — the dynamic library file to load. Repeat this option for every dynlib module instance added to the module-config option. 🧚‍♀️

🦋 DNS64 MODULE OPTIONS 🦋

These options are part of the server: section.

The dns64 module must be configured in the module-config directive, e.g.:

module-config: "dns64 validator iterator"

and be compiled into the daemon to be enabled.

NOTE: if combining the respip and dns64 modules, the respip module needs to appear before the dns64 module in the module-config configuration option so that response IP and/or RPZ feeds can properly filter responses regardless of DNS64 synthesis.

dns64-prefix: <IPv6 prefix> — this sets the DNS64 prefix to use to synthesize AAAA records with. It must be /96 or shorter.

Default: 64:ff9b::/96

dns64-synthall: <yes or no> — WARNING: Debugging feature! If enabled, synthesize all AAAA records despite the presence of actual AAAA records.

Default: no

dns64-ignore-aaaa: <domain name> — list domain for which the AAAA records are ignored and the A record is used by DNS64 processing instead. Can be entered multiple times, list a new domain for which it applies, one per line. Applies also to names underneath the name given. 💅

🌈 NAT64 OPTIONS 🌈

These options are part of the server: section.

NAT64 operation allows using a NAT64 prefix for outbound requests to IPv4-only servers.

do-nat64: <yes or no> — use NAT64 to reach IPv4-only servers. Consider also enabling prefer-ip6 to prefer native IPv6 connections to nameservers.

Default: no

nat64-prefix: <IPv6 prefix> — use a specific NAT64 prefix to reach IPv4-only servers. The prefix length must be one of /32, /40, /48, /56, /64 or /96.

The NAT64 prefix is allowed by the do-not-query-address option, so that there is a clear outcome of addresses in both; the NAT64 prefix is allowed. The IPv4 address could be filtered by the do-not-query-address option, if needed. Allowing the NAT64 prefix is useful when using do-not-query-address for a cluster of machines that is IPv6-only and uses NAT64, but does not have internet access.

Default: 64:ff9b::/96 (same as dns64-prefix) 🌸

💅 DNSCRYPT OPTIONS 💅

These options are part of the dnscrypt: section.

The dnscrypt: section gives the settings of the dnscrypt channel. While those options are available, they are only meaningful if Unbound was compiled with --enable-dnscrypt. Currently certificate and secret/public keys cannot be generated by Unbound. You can use dnscrypt-wrapper to generate those:

https://github.com/cofyc/dnscrypt-wrapper/blob/master/README.md#usage

dnscrypt-enable: <yes or no> — whether or not the dnscrypt config should be enabled. You may define configuration but not activate it.

Default: no

dnscrypt-port: <port number> — on which port should dnscrypt should be activated.

NOTE: there should be a matching interface option defined in the server: section for this port.

dnscrypt-provider: <provider name> — the provider name to use to distribute certificates. This is of the form:

2.dnscrypt-cert.example.com.

IMPORTANT: the name MUST end with a dot.

dnscrypt-secret-key: <path to secret key file> — path to the time limited secret key file. This option may be specified multiple times.
dnscrypt-provider-cert: <path to cert file> — path to the certificate related to the dnscrypt-secret-key. This option may be specified multiple times.
dnscrypt-provider-cert-rotated: <path to cert file> — path to a certificate that we should be able to serve existing connection from but do not want to advertise over dnscrypt-provider 's TXT record certs distribution.

A typical use case is when rotating certificates, existing clients may still use the client magic from the old cert in their queries until they fetch and update the new cert. Likewise, it would allow one to prime the new cert/key without distributing the new cert yet, this can be useful when using a network of servers using anycast and on which the configuration may not get updated at the exact same time.

By priming the cert, the servers can handle both old and new certs traffic while distributing only one.

This option may be specified multiple times. 🎀

dnscrypt-shared-secret-cache-size: <memory size> — give the size of the data structure in which the shared secret keys are kept in. In bytes or use m(mega), k(kilo), g(giga). The shared secret cache is used when a same client is making multiple queries using the same public key. It saves a substantial amount of CPU.

Default: 4m

dnscrypt-shared-secret-cache-slabs: <number> — number of slabs in the dnscrypt shared secrets cache. Slabs reduce lock contention by threads. Must be set to a power of 2. Setting (close) to the number of cpus is a fairly good setting. If left unconfigured, it will be configured automatically to be a power of 2 close to the number of configured threads in multi-threaded environments.

Default: (unconfigured)

dnscrypt-nonce-cache-size: <memory size> — give the size of the data structure in which the client nonces are kept in. In bytes or use m(mega), k(kilo), g(giga). The nonce cache is used to prevent dnscrypt message replaying. Client nonce should be unique for any pair of client pk/server sk.

Default: 4m

dnscrypt-nonce-cache-slabs: <number> — number of slabs in the dnscrypt nonce cache. Slabs reduce lock contention by threads. Must be set to a power of 2. Setting (close) to the number of cpus is a fairly good setting. If left unconfigured, it will be configured automatically to be a power of 2 close to the number of configured threads in multi-threaded environments.

Default: (unconfigured) 🔥

💅 EDNS CLIENT SUBNET MODULE OPTIONS 💅

These cuties live in the server: section, bestie. Just so you know.

The ECS module has to be configured in the module-config directive, like this:

module-config: "subnetcache validator iterator"

…and it also has to be compiled into the daemon before it'll actually do anything. Both, okay? Not one or the other.

So here's the vibe ✨ if the destination address is allowed in the configuration, Unbound will add the EDNS0 option to the query, and that option carries the relevant part of the client's address. And when an answer comes back with the ECS option in it? The response and the option get parked in a specialized cache. If the authority said "nah, no support," then the response goes in the regular cache instead. Simple as that 💁

Also — extra detail, don't skip — when a client includes the option in its own queries, Unbound will forward that option when it sends the query off to addresses that are explicitly allowed in the configuration using send-client-subnet. The option will always be forwarded, regardless of the allowed addresses, when client-subnet-always-forward: yes is set. And in that case, the lookup in the regular cache gets skipped. Yep. Skipped. 🙈

The max size of the ECS cache is governed by msg-cache-size in the config file. On top of that, for each query only 100 different subnets are allowed to be stored per address family. Go over that number and the older entries get purged from cache. Bye bestie 👋

Okay so real talk for a sec 🫶 — EDNS Client Subnet segregates the client IP space to try and get tailored responses for prefixes of unknown sizes, and because of how that works, resolution and cache response performance both take a hit. So only enable subnetcache module in installations that actually need this, where the resolver and the clients belong to different networks. An open resolver installation is the classic example of that. Don't just turn it on for the aesthetic, ngl.

Also important: this module does not interact with the serve-expired* and prefetch options. They're not friends, they don't talk. 💀

send-client-subnet: <IP address> — send the client source address to this authority. Append /num to indicate a classless delegation netblock, like 10.2.3.4/24 or 2001::11/64 for example. Can be given multiple times. Authorities that aren't listed will not receive edns-subnet information, unless the domain in the query is specified in client-subnet-zone.
client-subnet-zone: <domain> — send the client source address in queries for this domain and its subdomains. Can be given multiple times. Zones that aren't listed won't receive edns-subnet information, unless they're hosted by an authority specified in send-client-subnet.
client-subnet-always-forward: <yes or no> — specifies whether the ECS address check (configured with send-client-subnet) is applied for all queries — even if the triggering query contains an ECS record — or only for queries where the ECS record is generated from the querier address (meaning there was no ECS data in the client query to begin with). If enabled, the address check is skipped when the client query contains an ECS record. And the lookup in the regular cache is skipped too.

Default: no

max-client-subnet-ipv6: <number> — the maximum prefix length of the client source address that we're willing to expose to third parties, for IPv6.

Default: 56

max-client-subnet-ipv4: <number> — the maximum prefix length of the client source address that we're willing to expose to third parties, for IPv4.

Default: 24

min-client-subnet-ipv6: <number> — the minimum prefix length of the IPv6 source mask that we're willing to accept in queries. Shorter source masks get you REFUSED answers. A source mask of 0 is always accepted, no drama.

Default: 0

min-client-subnet-ipv4: <number> — the minimum prefix length of the IPv4 source mask that we're willing to accept in queries. Shorter source masks get you REFUSED answers. A source mask of 0 is always accepted.

Default: 0

max-ecs-tree-size-ipv4: <number> — the maximum number of subnets of ECS answers kept in the ECS radix tree. This number applies for each qname/qclass/qtype tuple. Like, individually. Per tuple. 👑

Default: 100

max-ecs-tree-size-ipv6: <number> — the maximum number of subnets of ECS answers kept in the ECS radix tree. This number applies for each qname/qclass/qtype tuple, same deal.

Default: 100

🔐 OPPORTUNISTIC IPSEC SUPPORT MODULE OPTIONS 🔐

These ones also live in the server: section, bestie.

The IPsec module must be configured in the module-config directive, e.g.:

module-config: "ipsecmod validator iterator"

…aaand it needs to be compiled into Unbound by using --enable-ipsecmod to be enabled. Two steps, always two steps. 💫

So when Unbound receives an A/AAAA query that is not in the cache and it finds a valid answer, it'll hold off on returning that answer and instead generate an IPSECKEY subquery for the same domain name. If an answer was found, Unbound calls an external hook, passing these arguments:

QNAME : Domain name of the A/AAAA and IPSECKEY query. In string format.

IPSECKEY TTL : TTL of the IPSECKEY RRset.

A/AAAA : String of space separated IP addresses present in the A/AAAA RRset. The IP addresses are in string format.

IPSECKEY : String of space separated IPSECKEY RDATA present in the IPSECKEY RRset. The IPSECKEY RDATA are in DNS presentation format.

Then — and only then — the A/AAAA answer gets cached and returned to the client. If the external hook was called, the TTL changes to make sure it doesn't surpass ipsecmod-max-ttl. No cap, that's the whole point 🎀

The exact same procedure is followed when prefetch: yes is set, but with one lil difference: the A/AAAA answer is given to the client before the hook is called. And ipsecmod-max-ttl makes sure the A/AAAA answer served from cache is still relevant for opportunistic IPsec. Iconic behaviour, honestly.

ipsecmod-enabled: <yes or no> — specifies whether the IPsec module is enabled or not. Heads up: the IPsec module still needs to be defined in the module-config directive regardless. This option just makes it easy to turn the module on/off without restarting/reloading Unbound. Convenient ✨

Default: yes

ipsecmod-hook: <filename> — specifies the external hook that Unbound will call with system(3). The file can be given as an absolute/relative path. It needs the proper permissions to actually be executable by the same user that runs Unbound. And it must be present when the IPsec module is defined in the module-config directive — no hook, no vibes.
ipsecmod-strict: <yes or no> — if enabled, Unbound requires the external hook to return a success value of 0. Fail to do that and Unbound replies with SERVFAIL. The A/AAAA answer also won't be cached. Strict means strict, bestie 😤

Default: no

ipsecmod-max-ttl: <seconds> — time to live maximum for A/AAAA cached records after calling the external hook.

Default: 3600

ipsecmod-ignore-bogus: <yes or no> — specifies Unbound's behaviour when the IPSECKEY answer is bogus. Set it to yes and the hook will be called and the A/AAAA answer returned to the client. Set it to no and the hook won't be called and the answer to the A/AAAA query will be SERVFAIL. Mainly used for testing, fyi 🧪

Default: no

ipsecmod-allow: <domain> — allow the IPsec module functionality for this domain, so the module logic actually gets executed. Can be given multiple times, for different domains. If the option isn't specified at all, all domains are treated as being allowed (that's the default). 🦋
ipsecmod-whitelist: <domain> — alternate syntax for ipsecmod-allow. Same thing, different spelling. Slay.

💾 CACHE DB MODULE OPTIONS 💾

Okay bestie, these options are part of the cachedb: section — yes, the whole section is giving “second-level cache but make it extra.” ✨

The Cache DB module has to be configured in the module-config directive, like this:

module-config: "validator cachedb iterator"

and it has to be compiled into the daemon with --enable-cachedb. No cap.

So here’s the vibe: if this module is enabled and configured, the specified backend database works as a second level cache. When Unbound cannot find an answer to a query in its built-in in-memory cache, it consults the specified backend. If it finds a valid answer in the backend, Unbound uses it to respond to the query without performing iterative DNS resolution. If Unbound cannot even find an answer in the backend, it resolves the query as usual, and stores the answer in the backend. It’s literally the “check the group chat before asking again” of DNS. 💅

This module interacts with the serve-expired-* options and will reply with expired data if Unbound is configured for that. Yes, expired data can still make an entrance. 🫶

If Unbound was built with --with-libhiredis on a system that has installed the hiredis C client library of Redis, then the redis backend can be used. This backend communicates with the specified Redis server over a TCP connection to store and retrieve cache data. It can be used as a persistent and/or shared cache backend. Highkey iconic for sharing cache vibes. 🦋

NOTE: Unbound never removes data stored in the Redis server, even if some data have expired in terms of DNS TTL or the Redis server has cached too much data; if necessary the Redis server must be configured to limit the cache size, preferably with some kind of least-recently-used eviction policy. So yeah, Redis cleanup is not Unbound’s job — administrator bestie has to handle that. 😭

Additionally, the redis-expire-records option can be used in order to set the relative DNS TTL of the message as timeout to the Redis records; keep in mind that some additional memory is used per key and that the expire information is stored as absolute Unix timestamps in Redis (computer time must be stable). Stable time is not optional here, fr fr. 🕰️

This backend uses synchronous communication with the Redis server based on the assumption that the communication is stable and sufficiently fast. The thread waiting for a response from the Redis server cannot handle other DNS queries. Although the backend has the ability to reconnect to the server when the connection is closed unexpectedly and there is a configurable timeout in case the server is overly slow or hangs up, these cases are assumed to be very rare. If connection close or timeout happens too often, Unbound will be effectively unusable with this backend. It’s the administrator’s responsibility to make the assumption hold. 👑 So like, keep an eye on that connection, bestie.

The cachedb: section gives custom settings of the cache DB module. Here come the options, one by one, no skipping. 🎀

backend: <backend name> — Specify the backend database name. The default database is the in-memory backend named testframe, which, as the name suggests, is not of any practical use. Depending on the build-time configuration, redis backend may also be used as described above. Default: testframe
secret-seed: "<secret string>" — Specify a seed to calculate a hash value from query information. This value will be used as the key of the corresponding answer for the backend database and can be customized if the hash should not be predictable operationally. If the backend database is shared by multiple Unbound instances, all instances must use the same secret seed. Default: "default" 🔐
cachedb-no-store: <yes or no> — If the backend should be read from, but not written to. This makes this instance not store dns messages in the backend. But if data is available it is retrieved. Read-only era, honestly. Default: no
cachedb-check-when-serve-expired: <yes or no> — If enabled, the cachedb is checked before an expired response is returned. When serve-expired is enabled, without serve-expired-client-timeout, it then does not immediately respond with an expired response from cache, but instead first checks the cachedb for valid contents, and if so returns it. If the cachedb also has no valid contents, the serve expired response is sent. If also serve-expired-client-timeout is enabled, the expired response is delayed until the timeout expires. Unless the lookup succeeds within the timeout. It’s giving “pls wait, maybe there’s fresher tea.” Default: yes

The following cachedb: options are specific to the redis backend. Redis girlies, this is your section. 🍓

redis-server-host: <server address or name> — The IP (either v6 or v4) address or domain name of the Redis server. In general an IP address should be specified as otherwise Unbound will have to resolve the name of the server every time it establishes a connection to the server. Default: 127.0.0.1
redis-server-port: <port number> — The TCP port number of the Redis server. Default: 6379
redis-server-path: <unix socket path> — The unix socket path to connect to the Redis server. Unix sockets may have better throughput than the IP address option. Default: "" (disabled)
redis-server-password: "<password>" — The Redis AUTH password to use for the Redis server. Only relevant if Redis is configured for client password authorisation. Default: "" (disabled) 🙈
redis-timeout: <msec> — The period until when Unbound waits for a response from the Redis server. If this timeout expires Unbound closes the connection, treats it as if the Redis server does not have the requested data, and will try to re-establish a new connection later. Default: 100
redis-command-timeout: <msec> — The timeout to use for Redis commands, in milliseconds. If 0, it uses the redis-timeout value. Default: 0
redis-connect-timeout: <msec> — The timeout to use for Redis connection set up, in milliseconds. If 0, it uses the redis-timeout value. Default: 0
redis-expire-records: <yes or no> — If Redis record expiration is enabled. If yes, Unbound sets timeout for Redis records so that Redis can evict keys that have expired automatically. If Unbound is configured with serve-expired and serve-expired-ttl: 0, this option is internally reverted to "no". Default: no

NOTE: Redis "SET ... EX" support is required for this option (Redis >= 2.6.12). Do not come for this option with ancient Redis, bestie. 💀

redis-logical-db: <logical database index> — The logical database in Redis to use. These are databases in the same Redis instance sharing the same configuration and persisted in the same RDB/AOF file. If unsure about using this option, Redis documentation (https://redis.io/commands/select/) suggests not to use a single Redis instance for multiple unrelated applications. The default database in Redis is 0 while other logical databases need to be explicitly SELECT'ed upon connecting. Default: 0
redis-replica-server-host: <server address or name> — The IP (either v6 or v4) address or domain name of the Redis server. In general an IP address should be specified as otherwise Unbound will have to resolve the name of the server every time it establishes a connection to the server.

This server is treated as a read-only replica server (https://redis.io/docs/management/replication/#read-only-replica). If specified, all Redis read commands will go to this replica server, while the write commands will go to the redis-server-host. Default: "" (disabled). 🌈

redis-replica-server-port: <port number> — The TCP port number of the Redis replica server. Default: 6379
redis-replica-server-path: <unix socket path> — The unix socket path to connect to the Redis replica server. Unix sockets may have better throughput than the IP address option. Default: "" (disabled)
redis-replica-server-password: "<password>" — The Redis AUTH password to use for the Redis server. Only relevant if Redis is configured for client password authorisation. Default: "" (disabled) 🔐
redis-replica-timeout: <msec> — The period until when Unbound waits for a response from the Redis replica server. If this timeout expires Unbound closes the connection, treats it as if the Redis server does not have the requested data, and will try to re-establish a new connection later. Default: 100
redis-replica-command-timeout: <msec> — The timeout to use for Redis replica commands, in milliseconds. If 0, it uses the redis-replica-timeout value. Default: 0
redis-replica-connect-timeout: <msec> — The timeout to use for Redis replica connection set up, in milliseconds. If 0, it uses the redis-replica-timeout value. Default: 0
redis-replica-logical-db: <logical database index> — Same as redis-logical-db but for the Redis replica server. Default: 0

✨ DNSTAP OPTIONS ✨

These options live in the dnstap: section, bestie. DNSTAP is literally a flexible, structured binary log format for DNS software. When it’s compiled in by using --enable-dnstap, you can enable it in the dnstap: section. That starts an extra thread (when compiled with threading) that writes the log info to the destination. If Unbound is compiled without threading, it does not spawn a thread — it connects per-process to the destination instead. iykyk 💅

dnstap-enable: <yes or no> — Whether dnstap is enabled. If yes, it connects to the DNSTAP server, and if any of the dnstap-log-..-messages: options is enabled, it sends logs for those messages to the server.

Default: no

dnstap-bidirectional: <yes or no> — Use frame streams in bidirectional mode to transfer DNSTAP messages. fancy stream behavior, no cap.

Default: yes

dnstap-socket-path: <file name> — Sets the unix socket file name for connecting to the server that’s listening on that socket.

Default: /etc/unbound/dnstap.sock

dnstap-ip: <IPaddress[@port]> — If "", the unix socket is used. If it’s set with an IP address (IPv4 or IPv6), that address is used to connect to the server.

Default: ""

dnstap-tls: <yes or no> — Set this to use TLS to connect to the server specified in dnstap-ip. If set to no, TCP is used to connect to the server.

Default: yes

dnstap-tls-server-name: <name of TLS authentication> — The TLS server name to authenticate the server with. Used when dnstap-tls: yes is set. If "", it’s ignored.

Default: ""

dnstap-tls-cert-bundle: <file name of cert bundle> — The pem file with certs to verify the TLS server certificate. If "", the server default cert bundle is used, or the windows cert bundle on windows.

Default: ""

dnstap-tls-client-key-file: <file name> — The client key file for TLS client authentication. If "", client authentication is not used.

Default: ""

dnstap-tls-client-cert-file: <file name> — The client cert file for TLS client authentication. certificate vibes.

Default: ""

dnstap-send-identity: <yes or no> — If enabled, the server identity is included in the log messages.

Default: no

dnstap-send-version: <yes or no> — If enabled, the server version is included in the log messages.

Default: no

dnstap-identity: <string> — The identity to send with messages. If "", the hostname is used.

Default: ""

dnstap-version: <string> — The version to send with messages. If "", the package version is used.

Default: ""

dnstap-sample-rate: <number> — The sample rate for log of messages: it logs only 1/N messages. With 0 it is disabled. This is useful in a high volume environment, where log functionality would otherwise not be reliable. For example, 10 would spend only 1/10th time on logging, and 100 would only spend a hundredth of the time on logging. math but make it logging ✨

Default: 0 (disabled)

dnstap-log-resolver-query-messages: <yes or no> — Enable to log resolver query messages. These are messages from Unbound to upstream servers.

Default: no

dnstap-log-resolver-response-messages: <yes or no> — Enable to log resolver response messages. These are replies from upstream servers to Unbound.

Default: no

dnstap-log-client-query-messages: <yes or no> — Enable to log client query messages. These are client queries to Unbound.

Default: no

dnstap-log-client-response-messages: <yes or no> — Enable to log client response messages. These are responses from Unbound to clients.

Default: no

dnstap-log-forwarder-query-messages: <yes or no> — Enable to log forwarder query messages. forwarder query era.

Default: no

dnstap-log-forwarder-response-messages: <yes or no> — Enable to log forwarder response messages. slay those forwarder responses into the log 🎀

Default: no

🛡️ RESPONSE POLICY ZONE OPTIONS 🛡️

okay bestie, these options live inside the rpz: section. Response Policy Zones are configured with rpz: section clauses, and every single one must have a name option. You can have multiple ones by listing multiple rpz: section clauses, each with a different name. RPZ sections are applied in order of configuration, and any match from an earlier RPZ zone will terminate the RPZ lookup. Also note: a PASSTHRU action is still considered a match. 💅

The respip module needs to be added to the module-config, like this:

module-config: "respip validator iterator"

NOTE: if you’re combining the respip and dns64 modules, the respip module needs to appear before the dns64 module in the module-config configuration option so that response IP and/or RPZ feeds can properly filter responses regardless of DNS64 synthesis. no cap, order matters here 🫶

QNAME, Response IP Address, nsdname, nsip and clientip triggers are supported. Supported actions are: NXDOMAIN, NODATA, PASSTHRU, DROP, Local Data, tcp-only and drop. RPZ QNAME triggers are applied after any local-zone and before any auth-zone. ✨

The RPZ zone is a regular DNS zone formatted with a SOA start record as usual. The items in the zone are entries that specify what to act on (the trigger) and what to do (the action). The trigger to act on is recorded in the name, and the action to do is recorded as the resource record. The names all end in the zone name, so you could type the trigger names without a trailing dot in the zonefile. 🎀

An example RPZ record that answers example.com with NXDOMAIN:

example.com CNAME .

The triggers are encoded in the name on the left:

name                          query name
netblock.rpz-client-ip        client IP address
netblock.rpz-ip               response IP address in the answer
name.rpz-nsdname              nameserver name
netblock.rpz-nsip             nameserver IP address

The netblock is written as <netblocklen>.<ip address in reverse>. For IPv6 use zz for ::. Specify individual addresses with scope length of 32 or 128. For example, 24.10.100.51.198.rpz-ip is 198.51.100.10/24 and 32.10.zz.db8.2001.rpz-ip is 2001:db8:0:0:0:0:0:10/32. 🦋

The actions are specified with the record on the right:

CNAME .                      nxdomain reply
CNAME *.                     nodata reply
CNAME rpz-passthru.          do nothing, allow to continue
CNAME rpz-drop.              the query is dropped
CNAME rpz-tcp-only.          answer over TCP
A 192.0.2.1                  answer with this IP address

Other records like AAAA, TXT and other CNAMEs (not rpz-..) can also be used to answer queries with that content. iconique. 👑

name: <zone name> — Name of the authority zone.
primary: <IP address or host name> — Where to download a copy of the zone from, with AXFR and IXFR. Multiple primaries can be specified. They are all tried if one fails.
- To use a non-default port for DNS communication, append '@' with the port number.
- You can append a '#' and a name, then AXFR over TLS can be used and the TLS authentication certificates will be checked with that name.
- If you combine the '@' and '#', the '@' comes first. If you point it at another Unbound instance, it would not work because that does not support AXFR/IXFR for the zone, but if you used url to download the zonefile as a text file from a webserver that would work.
- If you specify the hostname, you cannot use the domain from the zonefile, because it may not have that when retrieving that data; instead use a plain IP address to avoid a circular dependency on retrieving that IP address.
- Every number of IXFR transfers, a full AXFR is performed. This is to consolidate the rpz memory, that would otherwise grow. The fixed value is after 5 IXFR transfers.
master: <IP address or host name> — Alternate syntax for primary.
url: <url to zonefile> — Where to download a zonefile for the zone. With HTTP or HTTPS. An example for the url is:

http://www.example.com/example.org.zone

Multiple url statements can be given, they are tried in turn.

If only urls are given, the SOA refresh timer is used to wait for making new downloads. If also primaries are listed, the primaries are first probed with UDP SOA queries to see if the SOA serial number has changed, reducing the number of downloads. If none of the URLs work, the primaries are tried with IXFR and AXFR.
For HTTPS, the tls-cert-bundle and the hostname from the url are used to authenticate the connection.

allow-notify: <IP address or host name or netblockIP/prefix> — With allow-notify you can specify additional sources of notifies. When notified, the server attempts to first probe and then zone transfer. If the notify is from a primary, it first attempts that primary. Otherwise other primaries are attempted. If there are no primaries, but only urls, the file is downloaded when notified.
- NOTE: the primaries from primary and url statements are allowed notify by default.
zonefile: <filename> — The filename where the zone is stored. If not given then no zonefile is used. If the file does not exist or is empty, Unbound will attempt to fetch zone data (eg. from the primary servers).
rpz-action-override: <action> — Always use this RPZ action for matching triggers from this zone. Possible actions are: nxdomain, nodata, passthru, drop, disabled and cname.
rpz-cname-override: <domain> — The CNAME target domain to use if the cname action is configured for rpz-action-override.
rpz-log: <yes or no> — Log all applied RPZ actions for this RPZ zone. Default: no.
rpz-log-name: <name> — Specify a string to be part of the log line, for easy referencing.
rpz-signal-nxdomain-ra: <yes or no> — Signal when a query is blocked by the RPZ with NXDOMAIN with an unset RA flag. This allows certain clients, like dnsmasq, to infer that the domain is externally blocked. Default: no.
for-downstream: <yes or no> — If enabled, the zone is authoritatively answered for and queries for the RPZ zone information are answered to downstream clients. This is useful for monitoring scripts that can then access the SOA information to check if the RPZ information is up to date. Default: no.
tags: "<list of tags>" — Limit the policies from this RPZ section to clients with a matching tag. Tags need to be defined in define-tag and can be assigned to client addresses using access-control-tag or interface-tag. Enclose list of tags in quotes ("") and put spaces between tags. If no tags are specified, the policies from this section will be applied for all clients.
max-transfer-size: <number> — Number of bytes size of the maximum zone transfer size. Larger transfers, over AXFR, IXFR and HTTP, are not allowed. A plain number is in bytes; append 'k', 'm' or 'g' for kilobytes, megabytes or gigabytes (1024*1024 bytes in a megabyte). The value 0 disables the feature. Only consider for untrusted/misbehaving primaries that could hog resources and bring down the resolver. Default: 0.
max-transfer-time: <msec> — Maximum time in milliseconds that a zone transfer is allowed to take from the start. The value 0 disables the feature. Only consider for untrusted/misbehaving primaries that could hog resources and bring down the resolver. Default: 0.

🧠 MEMORY CONTROL EXAMPLE 🧠

In the example config settings below, memory usage is reduced. Some service levels are lower: notable very large data and a high TCP load are no longer supported. Very large data and high TCP loads are exceptional for the DNS. DNSSEC validation is enabled, just add trust anchors. If you do not have to worry about programs using more than 3 Mb of memory, the below example is not for you. Use the defaults to receive full service, which on BSD-32bit tops out at 30-40 Mb after heavy usage. 🫶

# example settings that reduce memory usage
server:
    num-threads: 1
    outgoing-num-tcp: 1 # this limits TCP service, uses less buffers.
    incoming-num-tcp: 1
    outgoing-range: 60  # uses less memory, but less performance.
    msg-buffer-size: 8192   # note this limits service, 'no huge stuff'.
    msg-cache-size: 100k
    msg-cache-slabs: 1
    rrset-cache-size: 100k
    rrset-cache-slabs: 1
    infra-cache-numhosts: 200
    infra-cache-slabs: 1
    key-cache-size: 100k
    key-cache-slabs: 1
    neg-cache-size: 10k
    num-queries-per-thread: 30
    target-fetch-policy: "2 1 0 0 0 0"
    harden-large-queries: "yes"
    harden-short-bufsize: "yes"

📁 FILES 📁

/etc/unbound — default Unbound working directory.
/etc/unbound — default chroot(2) location. yes, same path twice, that’s what it says bestie 💀
/etc/unbound/unbound.conf — Unbound configuration file.
/run/unbound.pid — default Unbound pidfile with process ID of the running daemon.
unbound.log — Unbound log file. Default is to log to syslog(3).

👀 SEE ALSO 👀

unbound(8), unbound-checkonf(8).

✍️ AUTHOR ✍️

Unbound developers are mentioned in the CREDITS file in the distribution.

1999-2026, NLnet Labs

okay that’s the whole vibe check bestie 💅✨ this page is from 1.26.2, dated Oct 21, 2026, and it’s the UNBOUND.CONF(5) manual page. slay 🎀🦋

📦 where this page comes from

arch rolling: unbound 1.26.2-1 · 0BSD AND BSD-3-Clause
💖 girlified by deepseek-ai/DeepSeek-V4.1-Flash-1017 on 2026-11-06: a language model's rewrite of the original, under its terms

UNBOUND.CONF(5)

💖 unbound.conf(5) — the Unbound 1.26.2 config file, bestie edition 🎀

✨ NAME ✨

🌸 SYNOPSIS 🌸

🦋 DESCRIPTION 🦋

💅 FILE FORMAT 💅

🌈 EXAMPLE 🌈

🎀 SECTION CLAUSES 🎀

🧚 INCLUDING FILES 🧚

👑 SERVER OPTIONS 👑

🚫 the local-zone types, explained

🏠 localhost

🔁 reverse IPv4 loopback

🔁 reverse IPv6 loopback

🏡 home.arpa (RFC 8375)

🔎 resolver.arpa (RFC 9462)

🛠️ service.arpa (draft-ietf-dnssd-srp-25)

🧅 onion (RFC 7686)

🧪 test (RFC 6761)

🚫 invalid (RFC 6761)

🏢 reverse local use zones (RFC 1918)

🌐 special-use IPv4 Addresses (RFC 3330)

💜 reverse IPv6 unspecified (RFC 4291)

🦋 reverse IPv6 Locally Assigned Local Addresses (RFC 4193)

🔗 reverse IPv6 Link Local Addresses (RFC 4291)

🌸 reverse IPv6 Example Prefix

✨ local-data: "<resource record string>" ✨

🌈 local-data-ptr: "IPaddr name" 🌈

🎀 local-zone-tag: <zone> <"list of tags"> 🎀

💫 local-zone-override: <zone> <IP netblock> <type> 💫

🔥 response-ip: <IP-netblock> <action> 🔥

🫶 response-ip-data: <IP-netblock> <"resource record string"> 🫶

🎛️ REMOTE CONTROL OPTIONS 🎛️

🦋 STUB ZONE OPTIONS 🦋

💜 FORWARD ZONE OPTIONS 💜

✨ AUTHORITY ZONE OPTIONS ✨

🌈 VIEW OPTIONS 🌈

💖 PYTHON MODULE OPTIONS 💖

🎀 DYNAMIC LIBRARY MODULE OPTIONS 🎀

🦋 DNS64 MODULE OPTIONS 🦋

🌈 NAT64 OPTIONS 🌈

💅 DNSCRYPT OPTIONS 💅

💅 EDNS CLIENT SUBNET MODULE OPTIONS 💅

🔐 OPPORTUNISTIC IPSEC SUPPORT MODULE OPTIONS 🔐

💾 CACHE DB MODULE OPTIONS 💾

✨ DNSTAP OPTIONS ✨

🛡️ RESPONSE POLICY ZONE OPTIONS 🛡️

🧠 MEMORY CONTROL EXAMPLE 🧠

📁 FILES 📁

👀 SEE ALSO 👀

✍️ AUTHOR ✍️

©️ COPYRIGHT ©️

💖 `unbound.conf(5)` — the Unbound 1.26.2 config file, bestie edition 🎀

🏠 `localhost`

🏡 `home.arpa` (RFC 8375)

🔎 `resolver.arpa` (RFC 9462)

🛠️ `service.arpa` (draft-ietf-dnssd-srp-25)

🧅 `onion` (RFC 7686)

🧪 `test` (RFC 6761)

🚫 `invalid` (RFC 6761)

✨ `local-data: "<resource record string>"` ✨

🌈 `local-data-ptr: "IPaddr name"` 🌈

🎀 `local-zone-tag: <zone> <"list of tags">` 🎀

💫 `local-zone-override: <zone> <IP netblock> <type>` 💫

🔥 `response-ip: <IP-netblock> <action>` 🔥

🫶 `response-ip-data: <IP-netblock> <"resource record string">` 🫶