Release Notes 

Introduction 

BIND 9.20 is a stable branch, suitable for production use. This document summarizes significant changes since the last production release on the 9.18 branch. Please see the Changelog file for a more detailed list of changes and bug fixes.

Supported Platforms 

See the Supported Platforms section in the Resource Requirements chapter.

Download 

The latest versions of BIND 9 software can always be found at https://www.isc.org/download/. There you will find additional information about each release, and source code.

Known Issues 

The list of known issues affecting the latest version in the 9.20 branch can be found at https://gitlab.isc.org/isc-projects/bind9/-/wikis/Known-Issues-in-BIND-9.20

Notes for BIND 9.20.4 

New Features 

Update built-in bind.keys file with the new 2025 IANA root key.

Add an initial-ds entry to bind.keys for the new root key, ID 38696, which is scheduled for publication in January 2025. [GL #4896]

Removed Features 

Move contributed DLZ modules into a separate repository. DLZ modules should not be used except in testing.

The DLZ modules were not maintained, the DLZ interface itself is going to be scheduled for removal, and the DLZ interface is blocking. Any module that blocks the query to the database blocks the whole server.

The DLZ modules now live in https://gitlab.isc.org/isc-projects/dlz-modules repository. [GL #4865]

Feature Changes 

dnssec-ksr now supports KSK rollovers.

The tool now allows for KSK generation, as well as planned KSK rollovers. When signing a bundle from a Key Signing Request (KSR), only the key that is active in that time frame is used for signing. Also, the CDS and CDNSKEY records are now added and removed at the correct time. [GL #4697] [GL #4705]
Print RFC 7314: EXPIRE option in transfer summary. [GL #5013]
Emit more helpful log messages for exceeding max-records-per-type.

The new log message is emitted when adding or updating an RRset fails due to exceeding the max-records-per-type limit. The log includes the owner name and type, corresponding zone name, and the limit value. It will be emitted on loading a zone file, inbound zone transfer (both AXFR and IXFR), handling a DDNS update, or updating a cache DB. It’s especially helpful in the case of zone transfer, since the secondary side doesn’t have direct access to the offending zone data.

It could also be used for max-types-per-name, but this change doesn’t implement it yet as it’s much less likely to happen in practice.
Harden key management when key files have become unavailable.

Prior to doing key management, BIND 9 will check if the key files on disk match the expected keys. If key files for previously observed keys have become unavailable, this will prevent the internal key manager from running.

Bug Fixes 

Use TLS for notifies if configured to do so.

Notifies configured to use TLS will now be sent over TLS, instead of plain text UDP or TCP. Also, failing to load the TLS configuration for notify now results in an error. [GL #4821]
{&dns} is as valid as {?dns} in a SVCB’s dohpath.

dig failed to parse a valid SVCB record with a dohpath URI template containing a {&dns}, like dohpath=/some/path?key=value{&dns}”. [GL #4922]
Fix NSEC3 closest encloser lookup for names with empty non-terminals.

A previous performance optimization for finding the NSEC3 closest encloser when generating authoritative responses could cause servers to return incorrect NSEC3 records in some cases. This has been fixed. [GL #4950]
recursive-clients statement with value 0 triggered an assertion failure.

BIND 9.20.0 broke recursive-clients 0;. This has now been fixed. [GL #4987]
Parsing of hostnames in rndc.conf was broken.

When DSCP support was removed, parsing of hostnames in rndc.conf was accidentally broken, resulting in an assertion failure. This has been fixed. [GL #4991]
dig options of the form [+-]option=<value> failed to display the value on the printed command line. This has been fixed. [GL #4993]
Provide more visibility into TLS configuration errors by logging SSL_CTX_use_certificate_chain_file() and SSL_CTX_use_PrivateKey_file() errors individually. [GL #5008]
Fix a race condition when canceling ADB find which could cause an assertion failure. [GL #5024]
SERVFAIL cache memory cleaning is now more aggressive; it no longer consumes a lot of memory if the server encounters many SERVFAILs at once. [GL #5025]
Fix trying the next primary XoT server when the previous one was marked as unreachable.

In some cases named failed to try the next primary server in the primaries list when the previous one was marked as unreachable. This has been fixed. [GL #5038]

Notes for BIND 9.20.3 

New Features 

Log query response status to the query log.

Log a query response summary using the new responses category. Logging can be controlled via the responselog option and via rndc responselog. [GL #459]
Added WALLET type.

Add the new record type WALLET (262). This provides a mapping from a domain name to a cryptographic currency wallet. Multiple mappings can exist if multiple records exist. [GL #4947]

Feature Changes 

Set logging category for notify/xfer-in-related messages.

Some notify and xfer-in-related log messages were logged at the “general” category level instead of their own category. This has been fixed. [GL #2730]
Allow IXFR-to-AXFR fallback on DNS_R_TOOMANYRECORDS.

This change allows fallback from an IXFR failure to AXFR when the reason is DNS_R_TOOMANYRECORDS. [GL #4928]

Bug Fixes 

Fix a statistics channel counter bug when “forward only” zones are used.

When resolving a zone with a “forward only” policy, and finding out that all the forwarders were marked as “bad”, the “ServerQuota” counter of the statistics channel was incorrectly increased. This has been fixed. [GL #1793]
Fix a bug in the static-stub implementation.

Static-stub addresses and addresses from other sources were being mixed together, resulting in static-stub queries going to addresses not specified in the configuration, or alternatively, static-stub addresses being used instead of the correct server addresses. [GL #4850]
Don’t allow statistics-channels if libxml2 and libjson-c are not configured.

When BIND 9 is not configured with the libxml2 and libjson-c libraries, the use of the statistics-channels option is a fatal error. [GL #4895]
Separate DNSSEC validation from long-running tasks.

Split CPU-intensive and long-running tasks into separate threadpools in a way that the long-running tasks - like RPZ, catalog zone processing, or zone file operations - don’t block CPU-intensive operations like DNSSEC validations. [GL #4898]
Fix an assertion failure when processing access control lists.

The named process could terminate unexpectedly when processing ACLs. This has been fixed. [GL #4908]
Fix a bug in Offline KSK using a ZSK with an unlimited lifetime.

If the ZSK had an unlimited lifetime, the timing metadata Inactive and Delete could not be found and were treated as an error, preventing the zone from being signed. This has been fixed. [GL #4914]
Limit the outgoing UDP send queue size.

If the operating system UDP queue got full and the outgoing UDP sending started to be delayed, BIND 9 could exhibit memory spikes as it tried to enqueue all the outgoing UDP messages. It now tries to deliver the outgoing UDP messages synchronously; if that fails, it drops the outgoing DNS message that would get queued up and then timeout on the client side. [GL #4930]
Do not set SO_INCOMING_CPU.

Remove the SO_INCOMING_CPU setting as kernel scheduling performs better without constraints. [GL #4936]
Fix the rndc dumpdb command’s error reporting.

The rndc dumpdb command was not reporting errors that occurred when named started up the database dump process. This has been fixed. [GL #4944]
Fix long-running incoming transfers.

Incoming transfers that took longer than 30 seconds would stop reading from the TCP stream and the incoming transfer would be indefinitely stuck, causing BIND 9 to hang during shutdown.

This has been fixed, and the max-transfer-time-in and max-transfer-idle-in timeouts are now honored. [GL #4949]
Fix an assertion failure when receiving DNS responses over TCP.

When matching the received Query ID in the TCP connection, an invalid Query ID could cause an assertion failure. This has been fixed. [GL #4952]

Known Issues 

There are no new known issues with this release. See above for a list of all known issues affecting this BIND 9 branch.

Notes for BIND 9.20.2 

New Features 

Support for Offline KSK implemented.

Add a new configuration option offline-ksk to enable Offline KSK key management. Signed Key Response (SKR) files created with dnssec-ksr (or other programs) can now be imported into named with the new rndc skr -import command. Rather than creating new DNSKEY, CDS, and CDNSKEY records and generating signatures covering these types, these records are loaded from the currently active bundle from the imported SKR.

The implementation is loosely based on draft-icann-dnssec-keymgmt-01.txt. [GL #1128]
Print the full path of the working directory in startup log messages.

named now prints its initial working directory during startup, and the changed working directory when loading or reloading its configuration file, if it has a valid directory option defined. [GL #4731]
Support a restricted key tag range when generating new keys.

When multiple signers are being used to sign a zone, it is useful to be able to specify a restricted range of key tags to be used by an operator to sign the zone. The range can be specified with tag-range in dnssec-policy’s keys (for named and dnssec-ksr) and with the new options dnssec-keyfromlabel -M and dnssec-keygen -M. [GL #4830]

Feature Changes 

Exempt prefetches from the fetches-per-zone and fetches-per-server quotas.

Fetches generated automatically as a result of prefetch are now exempt from the fetches-per-zone and fetches-per-server quotas. This should help in maintaining the cache from which query responses can be given. [GL #4219]
Improve performance for queries that require an NSEC3 wildcard proof.

Rather than starting from the longest matching part of the requested name, lookup the shortest partial match. Most of the time this will be the actual closest encloser. [GL #4460]
Follow the number of CPUs set by taskset/cpuset.

Administrators may wish to constrain the set of cores that named runs on via the taskset, cpuset, or numactl programs (or equivalents on other OSes).

If the admin has used taskset, named now automatically uses the given number of CPUs rather than the system-wide count. [GL #4884]

Bug Fixes 

Delay the release of root privileges until after configuring controls.

Delay relinquishing root privileges until the control channel has been configured, for the benefit of systems that require root to use privileged port numbers. This mostly affects systems without fine- grained privilege systems (i.e., other than Linux). [GL #4793]
Fix a rare assertion failure when shutting down incoming transfer.

A very rare assertion failure could be triggered when the incoming transfer was either forcefully shut down, or it finished during the printing of the details about the statistics channel. This has been fixed. [GL #4860]
Fix algorithm rollover bug when there are two keys with the same keytag.

If there was an algorithm rollover and two keys of different algorithms shared the same keytags, there was the possibility that the check of whether the key matched a specific state could be performed against the wrong key. This has been fixed by not only checking for the matching key tag but also the key algorithm. [GL #4878]
Fix an assertion failure in validate_dnskey_dsset_done().

Under rare circumstances, named could terminate unexpectedly when validating a DNSKEY resource record if the validation had been canceled in the meantime. This has been fixed. [GL #4911]

Known Issues 

Long-running tasks in offloaded threads (e.g. the loading of RPZ zones or processing zone transfers) may block the resolution of queries during these operations and cause the queries to time out.

To work around the issue, the UV_THREADPOOL_SIZE environment variable can be set to a larger value before starting named. The recommended value is the number of RPZ zones (or number of transfers) plus the number of threads BIND should use, which is typically the number of CPUs. [GL #4898]

Notes for BIND 9.20.1 

New Features 

Implement rndc retransfer -force.

A new optional argument -force has been added to the command rndc retransfer. When it is specified, named aborts the ongoing zone transfer (if there is one) and starts a new transfer. [GL #2299] [GL !9219]
dig now reports a missing QUESTION section for messages with opcode QUERY.

Query responses should contain the QUESTION section, with some exceptions. dig was not reporting this. [GL #4808] [GL !9269]

Feature Changes 

Tighten max-recursion-queries and add max-query-restarts configuration statement.

There were cases when the max-recursion-queries quota was ineffective. It was possible to craft zones that would cause a resolver to waste resources by sending excessive queries while attempting to resolve a name. This has been addressed by correcting errors in the implementation of max-recursion-queries and by reducing the default value from 100 to 32.

In addition, a new max-query-restarts configuration statement has been added, which limits the number of times a recursive server will follow CNAME or DNAME records before terminating resolution. This was previously a hard-coded limit of 16 but is now configurable with a default value of 11.

ISC would like to thank Huayi Duan, Marco Bearzi, Jodok Vieli, and Cagin Tanir from NetSec group, ETH Zurich for discovering and notifying us about the issue. [GL #4741] [GL !9282]
Allow shorter resolver-query-timeout configuration.

The minimum allowed value of resolver-query-timeout was lowered from its previous value of 10 000 milliseconds (which is still the default) to 301 milliseconds. Note however that values of 1 to 300 inclusive are interpreted as seconds before applying the limit. A value of zero is interpreted as the default. [GL #4320] [GL !9220]
Raise the log level of priming failures.

When a priming query is complete, it was previously logged at level DEBUG(1), regardless of success or failure. It is now logged to NOTICE in the case of failure. [GL #3516] [GL !9250]

Bug Fixes 

Fix a crash caused by valid TSIG signatures with invalid time.

An assertion failure was triggered when the TSIG had a valid cryptographic signature but the time was invalid. This could happen when the times between the primary and secondary servers were not synchronised. The crash has now been fixed. [GL #4811] [GL !9245]
Return SERVFAIL for a too long CNAME chain.

When following long CNAME chains, named was returning NOERROR (along with a partial answer) instead of SERVFAIL, if the chain exceeded the maximum length. This has been fixed. [GL #4449] [GL !9203]
Reconfigure catz member zones during named reconfiguration.

During a reconfiguration, named wasn’t reconfiguring catalog zones’ member zones. This has been fixed. [GL #4733]
Update key lifetime and metadata after dnssec-policy reconfiguration.

Adjust key state and timing metadata if dnssec-policy key lifetime configuration is updated, so that it also affects existing keys. [GL #4677] [GL !9191]
Fix a crash during zone modification.

Fix an assertion failure that could happen when an authoritative zone was modified while the server was generating an answer from that zone. [GL #4691] [GL !9126]
Fix assertion failure when executing named-checkconf -v to print its version. [GL #4827] [GL !9246]
Fix generation of 6to4-self name expansion from IPv4 address.

The period between the most significant nibble of the encoded IPv4 address and the 2.0.0.2.IP6.ARPA suffix was missing, resulting in the wrong name being checked. This has been fixed. [GL #4766] [GL !9217]
dig +yaml was producing unexpected and/or invalid YAML. output. [GL #4796] [GL !9213]
SVBC ALPN text parsing failed to reject zero-length ALPN. [GL #4775] [GL !9209]
Fix false QNAME minimisation error being reported.

Remove the false positive success resolving log message when QNAME minimisation is in effect and the final result is an NXDOMAIN. [GL #4784] [GL !9215]
Fix --enable-tracing build on systems without dtrace.

A missing util/dtrace.sh file prevented builds on systems without the dtrace utility. This has been corrected. [GL #4835] [GL !9272]

Known Issues 

There are no new known issues with this release. See above for a list of all known issues affecting this BIND 9 branch.

Notes for BIND 9.20.0 

Note

This section only lists changes since BIND 9.18.28, the most recent release on the previous stable branch of BIND at the time of the publication of BIND 9.20.0.

New Features 

The forwarders statement now supports the tls argument, to be used to forward queries to DoT-enabled servers. [GL #3726]
named now supports forwarding Dynamic DNS updates through DNS-over-TLS (DoT). [GL #3512]
The nsupdate tool now supports DNS-over-TLS (DoT). [GL !6752]
The tls block was extended with a new cipher-suites option that allows permitted cipher suites for TLSv1.3 to be set. Please consult the documentation for additional details. [GL #3504]
Initial support for the PROXYv2 protocol was added. named can now accept PROXYv2 headers over all currently implemented DNS transports and dig can insert these headers into the queries it sends. Please consult the related documentation (allow-proxy, allow-proxy-on, listen-on, and listen-on-v6 for named, dig +proxy and dig +proxy-plain for dig) for additional details. [GL #4388]
The client-side support of the EDNS EXPIRE option has been expanded to include IXFR and AXFR query types. This enhancement enables named to perform AXFR and IXFR queries while incorporating the EDNS EXPIRE option. [GL #4170]
A new configuration option require-cookie has been introduced. It specifies whether there should be a DNS COOKIE in the response for a given prefix; if not, named falls back to TCP. This is useful if it is known that a given server supports DNS COOKIE. It can also be used to force all non-DNS COOKIE responses to fall back to TCP. [GL #2295]
The check-svcb option has been added to control the checking of additional constraints on SVCB records. This change affects named, named-checkconf, named-checkzone, named-compilezone, and nsupdate. [GL #3576]
The new resolver-use-dns64 option enables named to apply dns64 rules to IPv4 server addresses when sending recursive queries, so that resolution can be performed over a NAT64 connection. [GL #608]
A new option to dnssec-policy has been added, cdnskey, that allows users to enable or disable the publication of CDNSKEY records. [GL #4050]
When using dnssec-policy, it is now possible to configure the digest type to use when CDS records need to be published with cds-digest-types. Also, publication of specific CDNSKEY/CDS records can now be set with dnssec-signzone -G. [GL #3837]
Support for multi-signer model 2 (RFC 8901) when using inline-signing was added. [GL #2710]
HSM support was added to dnssec-policy. Keys can now be configured with a key-store that allows users to set the directory where key files are stored and to set a PKCS#11 URI string. The latter requires OpenSSL 3 and a valid PKCS#11 provider to be configured for OpenSSL. [GL #1129]
A new DNSSEC tool dnssec-ksr has been added to create Key Signing Request (KSR) and Signed Key Response (SKR) files. [GL #1128]
dnssec-verify and dnssec-signzone now accept a -J option to specify a journal file to read when loading the zone to be verified or signed. [GL #2486]
dnssec-keygen now allows the options -k and -f to be used together. This allows the creation of keys for a given dnssec-policy that match only the KSK (-fK) or ZSK (-fZ) roles. [GL #1128]
The response-policy statement was extended with a new argument ede. It enables an RFC 8914 Extended DNS Error (EDE) code of choice to be set for responses which have been modified by a given RPZ. [GL #3410]
A new way of configuring the preferred source address when talking to remote servers, such as primaries and parental-agents, has been added: setting the source and/or source-v6 arguments for a given statement is now possible. This new approach is intended to eventually replace statements such as parental-source, parental-source-v6, transfer-source, etc. [GL #3762]
The new command-line delv +ns option activates name server mode, to more accurately reproduce the behavior of named when resolving a query. In this mode, delv uses an internal recursive resolver rather than an external server. All messages sent and received during the resolution and validation process are logged. This can be used in place of dig +trace. [GL #3842]
The read timeout in rndc can now be specified on the command line using the -t option, allowing commands that take a long time to complete sufficient time to do so. [GL #4046]
The statistics channel now includes information about incoming zone transfers that are currently in progress. [GL #3883]
Information on incoming zone transfers in the statistics channel now also shows the zones’ “first refresh” flag, which indicates that a zone is not fully ready and that its first ever refresh is pending or is in progress. The number of such zones is now also exposed by the rndc status command. [GL #4241]
Added a new statistics variable recursive high-water that reports the maximum number of simultaneous recursive clients BIND has handled while running. [GL #4668]
A new command, rndc fetchlimit, prints a list of name server addresses that are currently rate-limited due to fetches-per-server and domain names that are rate-limited due to fetches-per-zone. [GL #665]
Queries and responses now emit distinct dnstap entries for DNS-over-TLS (DoT) and DNS-over-HTTPS (DoH), and dnstap-read understands these entries. [GL #4523]
dnstap-read can now print long timestamps with millisecond precision. [GL #2360]
Support for libsystemd’s sd_notify() function was added, enabling named to report its status to the init system. This allows systemd to wait until named is fully ready before starting other services that depend on name resolution. [GL #1176]
Support for User Statically Defined Tracing (USDT) probes has been added. These probes enable fine-grained application tracing and introduce no overhead when they are not enabled. [GL #4041]

Removed Features 

Support for Red Hat Enterprise Linux version 7 (and clones) has been dropped. A C11-compliant compiler is now required to compile BIND 9. [GL #3729]
Compiling with jemalloc versions older than 4.0.0 is no longer supported; those versions do not provide the features required by current BIND 9 releases. [GL #4296]
The auto-dnssec configuration statement has been removed. Please use dnssec-policy or manual signing instead. See article how to migrate from auto-dnssec to dnssec-policy.

The following statements have become obsolete: dnskey-sig-validity, dnssec-dnskey-kskonly, dnssec-update-mode, sig-validity-interval, and update-check-ksk. [GL #3672]
Dynamic updates that add and remove DNSKEY and NSEC3PARAM records no longer trigger key rollovers and denial-of-existence operations. This also means that the dnssec-secure-to-insecure option has been obsoleted. [GL #3686]
The glue-cache option has been removed. The glue cache feature still works and is now permanently enabled. [GL #2147]
Configuring the control channel to use a Unix domain socket has been a fatal error since BIND 9.18. The feature has now been completely removed and named-checkconf now reports it as a configuration error. [GL #4311]
The statements setting alternate local addresses for inbound zone transfers (alt-transfer-source, alt-transfer-source-v6, and use-alt-transfer-source) have been removed. [GL #3714]
The resolver-nonbackoff-tries and resolver-retry-interval statements have been removed. Using them is now a fatal error. [GL #4405]
BIND 9 no longer supports non-zero stale-answer-client-timeout values, when the feature is turned on. When using a non-zero value, named now generates a warning log message, and treats the value as 0. [GL #4447]
The Differentiated Services Code Point (DSCP) feature has been removed: configuring DSCP values in named.conf is now a configuration error. [GL #3789]
The keep-response-order option has been declared obsolete and the functionality has been removed. named expects DNS clients to be fully compliant with RFC 7766. [GL #3140]
Zone type delegation-only, and the delegation-only and root-delegation-only statements, have been removed. Using them is a configuration error.

These statements were created to address the SiteFinder controversy, in which certain top-level domains redirected misspelled queries to other sites instead of returning NXDOMAIN responses. Since top-level domains are now DNSSEC-signed, and DNSSEC validation is active by default, the statements are no longer needed. [GL #3953]
The coresize, datasize, files, and stacksize options have been removed. The limits these options set should be enforced externally, either by manual configuration (e.g. using ulimit) or via the process supervisor (e.g. systemd). [GL #3676]
Support for using AES as the DNS COOKIE algorithm (cookie-algorithm aes;) has been removed. The only supported DNS COOKIE algorithm is now the current default, SipHash-2-4. [GL #4421]
The TKEY Mode 2 (Diffie-Hellman Exchanged Keying Mode) has been removed and using TKEY Mode 2 is now a fatal error. Users are advised to switch to TKEY Mode 3 (GSS-API). [GL #3905]
Special-case code that was originally added to allow GSS-TSIG to work around bugs in the Windows 2000 version of Active Directory has now been removed, since Windows 2000 is long past end-of-life. The -o option and the oldgsstsig command to nsupdate have been deprecated, and are now treated as synonyms for -g and gsstsig respectively. [GL #4012]
Support for the lock-file statement and the named -X command-line option has been removed. An external process supervisor should be used instead. [GL #4391]

Alternatively, the flock utility (part of util-linux) can be used on Linux systems to achieve the same effect as lock-file or named -X:
```
flock -n -x <directory>/named.lock <path>/named <arguments>
```
The named command-line option -U, which specified the number of UDP dispatches, has been removed. Using it now returns a warning. [GL #1879]
The --with-tuning option for configure has been removed. Each of the compile-time settings that required different values based on the “workload” (which were previously affected by the value of the --with-tuning option) has either been removed or changed to a sensible default. [GL #3664]
The functions that were in the libbind9 shared library have been moved to the libisc and libisccfg libraries. The now-empty libbind9 has been removed and is no longer installed. [GL #3903]
The irs_resconf module has been moved to the libdns shared library. The now-empty libirs library has been removed and is no longer installed. [GL #3904]

Deprecated Features 

Features listed in this section still work but are scheduled for eventual removal.

The use of the max-zone-ttl option in options and zone blocks has been deprecated; it should now be configured as part of dnssec-policy. A warning is logged if this option is used in options or zone blocks. In a future release, it will become nonoperational. [GL #2918]
The sortlist option has been deprecated and will be removed in a future BIND 9.21.x release. Users should not rely on a specific order of resource records in DNS messages. [GL #4593]
The fixed value for the rrset-order option and the corresponding configure script option have been deprecated and will be removed in a future BIND 9.21.x release. Users should not rely on a specific order of resource records in DNS messages. [GL #4446]

Feature Changes 

BIND now depends on liburcu, Userspace RCU, for lock-free data structures. [GL #3934]
On Linux, libcap is now a required dependency to help named keep needed privileges. [GL #3583]
Compiling BIND 9 now requires at least libuv version 1.34.0 or higher. libuv should be available on all supported platforms either as a native package or as a backport. [GL #3567]
Outgoing zone transfers are no longer enabled by default. An explicit allow-transfer ACL must now be set at the zone, view, or options level to enable outgoing transfers. [GL #4728]
DNS zones signed using dnssec-policy now automatically detect their parent servers, and BIND queries them to check the content of the DS RRset. This allows DNSSEC key rollovers to safely and automatically proceed when the parent zone is updated with new DNSSEC keys, i.e. using the CDS/CDNSKEY mechanism. This behavior is facilitated by the new checkds feature, which automatically populates parental-agents by resolving the parent NS records. These parent name servers are queried to check the DS RRset during a KSK rollover initiated by dnssec-policy. [GL #3901]
The responsiveness of named was improved, when serving as an authoritative DNS server for a delegation-heavy zone(s) shortly after loading such zone(s). [GL #4045]
To improve query-processing latency under load, the uninterrupted time spent on resolving long chains of cached domain names has been reduced. [GL #4185]
QNAME minimization is now used when looking up the addresses of name servers during the recursive resolution process. [GL #4209]
BIND now returns BADCOOKIE for out-of-date or otherwise bad but well-formed DNS server cookies. [GL #4194]
The DNS name compression algorithm used in BIND 9 has been revised: it now compresses more thoroughly than before, so responses containing names with many labels might have a smaller encoding than before. [GL #3661]
Processing large incremental transfers (IXFR) has been offloaded to a separate work thread so that it does not prevent networking threads from processing regular traffic in the meantime. [GL #4367]
Querying the statistics channel no longer blocks DNS communication on the networking event loop level. [GL #4680]
The inline-signing zone option is now ignored if there is no dnssec-policy configured for the zone. This means that unsigned zones no longer create redundant signed versions of the zone. [GL #4349]
The inline-signing statement can now also be set inside dnssec-policy. The default is to use inline-signing. This also applies to the built-in policies default` and ``insecure. If inline-signing is set at the zone level, it overrides the value set in dnssec-policy. [GL #3677]
Due to the change in default value from no to yes, DNSSEC-enabled dynamic zones that do not have inline-signing explicitly set must now add the option to their configuration with the value no if they do not want their zone also to be inline-signed.
Following RFC 9276 recommendations, dnssec-policy now only allows an NSEC3 iteration count of 0 for the DNSSEC-signed zones using NSEC3 that the policy manages. [GL #4363]
The maximum number of NSEC3 iterations allowed for validation purposes has been lowered from 150 to 50. DNSSEC responses containing NSEC3 records with iteration counts greater than 50 are now treated as insecure. [GL #4363]
The dnssec-validation yes option now requires an explicitly configured trust-anchors statement. If using manual trust anchors is not operationally required, then please consider using dnssec-validation auto instead. [GL #4373]
named-compilezone no longer performs zone integrity checks by default; this allows faster conversion of a zone file from one format to another. [GL #4364]

Zone checks can be performed by running named-checkzone separately, or the previous default behavior can be restored by using:
```
named-compilezone -i full -k fail -n fail -r warn -m warn -M warn -S warn -T warn -W warn -C check-svcb:fail
```
The red-black tree data structure used in the RBTDB (the default database implementation for cache and zone databases), has been replaced with QP-tries. This is expected to improve performance and scalability, though in the current implementation large zones require roughly 15% more memory than the old red-black tree data structure.

A side effect of this change is that zone files that are created with masterfile-style relative - for example, the output of dnssec-signzone - will no longer have multiple different $ORIGIN statements. There should be no other changes to server behavior.

The old RBT-based database still exists for now, and can be used by specifying database rbt in a zone statement in named.conf, or by compiling with configure --with-zonedb=rbt --with-cachedb=rbt. [GL #4411] [GL #4614]
Multiple RNDC messages are now processed when sent in a single TCP message.

ISC would like to thank Dominik Thalhammer for reporting the issue and preparing the initial patch. [GL #4416]
The DNSSEC signing data included in zone statistics identified keys only by the key ID; this caused confusion when two keys using different algorithms had the same ID. Zone statistics now identify keys using the algorithm number, followed by “+”, followed by the key ID: for example, 8+54274. [GL #3525]
The TTL of the NSEC3PARAM record for every NSEC3-signed zone was previously set to 0. It is now changed to match the SOA MINIMUM value for the given zone. [GL #3570]
On startup, named now sets the limit on the number of open files to the maximum allowed by the operating system, instead of trying to set it to “unlimited”. [GL #3676]
When an international domain name is not valid according to IDNA2008, dig now tries to convert it according to IDNA2003 rules, or pass it through unchanged, instead of stopping with an error message. The idna2 utility can be used to check IDNA syntax. [GL #3527]
The memory statistics have been reduced to a single counter, InUse; Malloced is an alias that holds the same value. The other counters were usable with the old BIND 9 internal memory allocator, but they are unnecessary now that the latter has been removed. [GL #3718]
The log message resolver priming query complete has been moved from the INFO log level to the DEBUG(1) log level, to prevent delv from emitting that message when setting up its internal resolver. [GL #3842]
Worker threads’ event loops are now managed by a new “loop manager” API, significantly changing the architecture of the task, timer, and networking subsystems for improved performance and code flow. [GL #3508]
The code for DNS over TCP and DNS over TLS transports has been replaced with a new, unified transport implementation. [GL #3374]

Bug Fixes 

When the same notify-source address and port number was configured for multiple destinations and zones, an unresponsive server could tie up the relevant network socket until it timed out; in the meantime, NOTIFY messages for other servers silently failed. named will now retry sending such NOTIFY messages over TCP. Furthermore, NOTIFY failures are now logged at the INFO level. [GL #4001] [GL #4002]
DNS compression is no longer applied to the root name (.) if it is repeatedly used in the same RRset. [GL #3423]
named could incorrectly return non-truncated, glueless referrals for responses whose size was close to the UDP packet size limit. This has been fixed. [GL #1967]

Known Issues 

On some platforms, including FreeBSD, named must be run as root to use the rndc control channel on a privileged port (i.e., with a port number less than 1024; this includes the default rndc port, 953). Currently, using the named -u option to switch to an unprivileged user makes rndc unusable. This will be fixed in a future release; in the meantime, mac_portacl can be used as a workaround, as documented in https://kb.isc.org/docs/aa-00621. [GL #4793]
See above for a list of all known issues affecting this BIND 9 branch.

License 

BIND 9 is open source software licensed under the terms of the Mozilla Public License, version 2.0 (see the COPYING file for the full text).

Those wishing to discuss license compliance may contact ISC at https://www.isc.org/contact/.

End of Life 

BIND 9.20 is a stable branch, suitable for production use. After it has been in production use for a while it will be designated as an Extended Support Version (ESV). Until then, the current ESV is BIND 9.18, which will be supported until at least December 2025. See https://kb.isc.org/docs/aa-00896 for details of ISC’s software support policy.

Thank You 

Thank you to everyone who assisted us in making this release possible.