Release Notes

Task manager and socket code have been substantially modified. The manager uses per-cpu queues for tasks and network stack runs multiple event loops in CPU-affinitive threads. This greatly improves performance on large systems, especially when using multi-queue NICs.

Support for QNAME minimization was added and enabled by default in relaxed mode, in which BIND will fall back to normal resolution if the remote server returns something unexpected during the query minimization process. This default setting might change to strict in the future.

A new plugin mechanism has been added to allow extension of query processing functionality through the use of external libraries. The new filter-aaaa.so plugin replaces the filter-aaaa feature that was formerly implemented as a native part of BIND.

The plugin API is a work in progress and is likely to evolve as further plugins are implemented. [GL #15]

A new secondary zone option, mirror, enables named to serve a transferred copy of a zone's contents without acting as an authority for the zone. A zone must be fully validated against an active trust anchor before it can be used as a mirror zone. DNS responses from mirror zones do not set the AA bit ("authoritative answer"), but do set the AD bit ("authenticated data"). This feature is meant to facilitate deployment of a local copy of the root zone, as described in RFC 7706. [GL #33]

BIND now can be compiled against the libidn2 library to add IDNA2008 support. Previously, BIND supported IDNA2003 using the (now obsolete and unsupported) idnkit-1 library.

named now supports the "root key sentinel" mechanism. This enables validating resolvers to indicate which trust anchors are configured for the root, so that information about root key rollover status can be gathered. To disable this feature, add root-key-sentinel no; to named.conf. [GL #37]

The dnskey-sig-validity option allows the sig-validity-interval to be overriden for signatures covering DNSKEY RRsets. [GL #145]

When built on Linux, BIND now requires the libcap library to set process privileges. The adds a new compile-time dependency, which can be met on most Linux platforms by installing the libcap-dev or libcap-devel package. BIND can also be built without capability support by using configure --disable-linux-caps, at the cost of some loss of security.

The validate-except option specifies a list of domains beneath which DNSSEC validation should not be performed, regardless of whether a trust anchor has been configured above them. [GL #237]

Two new update policy rule types have been added krb5-selfsub and ms-selfsub which allow machines with Kerberos principals to update the name space at or below the machine names identified in the respective principals.

The new configure option --enable-fips-mode can be used to make BIND enable and enforce FIPS mode in the OpenSSL library. When compiled with such option the BIND will refuse to run if FIPS mode can't be enabled, thus this option must be only enabled for the systems where FIPS mode is available.

Two new configuration options min-cache-ttl and min-ncache-ttl has been added to allow the BIND 9 administrator to override the minimum TTL in the received DNS records (positive caching) and for storing the information about non-existent records (negative caching). The configured minimum TTL for both configuration options cannot exceed 90 seconds.

rndc status output now includes a reconfig/reload in progress status line if named configuration is being reloaded.

The new answer-cookie option, if set to no, prevents named from returning a DNS COOKIE option to a client, even if such an option was present in the request. This is only intended as a temporary measure, for use when named shares an IP address with other servers that do not yet support DNS COOKIE. A mismatch between servers on the same address is not expected to cause operational problems, but the option to disable COOKIE responses so that all servers have the same behavior is provided out of an abundance of caution. DNS COOKIE is an important security mechanism, and this option should not be used to disable it unless absolutely necessary.

Workarounds for servers that misbehave when queried with EDNS have been removed, because these broken servers and the workarounds for their noncompliance cause unnecessary delays, increase code complexity, and prevent deployment of new DNS features. See https://dnsflagday.net for further details.

In particular, resolution will no longer fall back to plain DNS when there was no response from an authoritative server. This will cause some domains to become non-resolvable without manual intervention. In these cases, resolution can be restored by adding server clauses for the offending servers, specifying edns no or send-cookie no, depending on the specific noncompliance.

To determine which server clause to use, run the following commands to send queries to the authoritative servers for the broken domain:

If the first command fails but the second succeeds, the server most likely needs send-cookie no. If the first two fail but the third succeeds, then the server needs EDNS to be fully disabled with edns no.

Please contact the administrators of noncompliant domains and encourage them to upgrade their broken DNS servers. [GL #150]

Previously, it was possible to build BIND without thread support for old architectures and systems without threads support. BIND now requires threading support (either POSIX or Windows) from the operating system, and it cannot be built without threads.

The filter-aaaa, filter-aaaa-on-v4, and filter-aaaa-on-v6 options have been removed from named, and can no longer be configured using native named.conf syntax. However, loading the new filter-aaaa.so plugin and setting its parameters provides identical functionality.

named can no longer use the EDNS CLIENT-SUBNET option for view selection. In its existing form, the authoritative ECS feature was not fully RFC-compliant, and could not realistically have been deployed in production for an authoritative server; its only practical use was for testing and experimentation. In the interest of code simplification, this feature has now been removed.

The ECS option is still supported in dig and mdig via the +subnet argument, and can be parsed and logged when received by named, but it is no longer used for ACL processing. The geoip-use-ecs option is now obsolete; a warning will be logged if it is used in named.conf. ecs tags in an ACL definition are also obsolete, and will cause the configuration to fail to load if they are used. [GL #32]

dnssec-keygen can no longer generate HMAC keys for TSIG authentication. Use tsig-keygen to generate these keys. [RT #46404]

Support for OpenSSL 0.9.x has been removed. OpenSSL version 1.0.0 or greater, or LibreSSL is now required.

The configure --enable-seccomp option, which formerly turned on system-call filtering on Linux, has been removed. [GL #93]

IPv4 addresses in forms other than dotted-quad are no longer accepted in master files. [GL #13] [GL #56]

IDNA2003 support via (bundled) idnkit-1.0 has been removed.

The "rbtdb64" database implementation (a parallel implementation of "rbt") has been removed. [GL #217]

The -r randomdev option to explicitly select random device has been removed from the ddns-confgen, rndc-confgen, nsupdate, dnssec-confgen, and dnssec-signzone commands.

The -p option to use pseudo-random data has been removed from the dnssec-signzone command.

Support for the RSAMD5 algorithm has been removed freom BIND as the usage of the RSAMD5 algorithm for DNSSEC has been deprecated in RFC6725, the security of the MD5 algorithm has been compromised, and its usage is considered harmful.

Support for the ECC-GOST (GOST R 34.11-94) algorithm has been removed from BIND, as the algorithm has been superseded by GOST R 34.11-2012 in RFC6986 and it must not be used in new deployments. BIND will neither create new DNSSEC keys, signatures and digests, nor it will validate them.

Support for DSA and DSA-NSEC3-SHA1 algorithms has been removed from BIND as the DSA key length is limited to 1024 bits and this is not considered secure enough.

named will no longer ignore "no-change" deltas when processing an IXFR stream. This had previously been permitted for compatibility with BIND 8, but now "no-change" deltas will trigger a fallback to AXFR as the recovery mechanism.

BIND 9 will no longer build on platforms that don't have proper IPv6 support. BIND 9 now also requires POSIX-compatible pthread support. Most of the platforms that lack these featuers are long past their end-of-lifew dates, and they are neither developed nor supported by their respective vendors.

The incomplete support for internationalization message catalogs has been removed from BIND. Since the internationalization was never completed, and no localized message catalogs were ever made available for the portions of BIND in which they could have been used, this change will have no effect except to simplify the source code. BIND's log messages and other output were already only available in English.

BIND will now always use the best CSPRNG (cryptographically-secure pseudo-random number generator) available on the platform where it is compiled. It will use the arc4random() family of functions on BSD operating systems, getrandom() on Linux and Solaris, CryptGenRandom on Windows, and the selected cryptography provider library (OpenSSL or PKCS#11) as the last resort. [GL #221]

The default setting for dnssec-validation is now auto, which activates DNSSEC validation using the IANA root key. (The default can be changed back to yes, which activates DNSSEC validation only when keys are explicitly configured in named.conf, by building BIND with configure --disable-auto-validation.) [GL #30]

BIND can no longer be built without DNSSEC support. A cryptography provider (i.e., OpenSSL or a hardware service module with PKCS#11 support) must be available. [GL #244]

Zone types primary and secondary are now available as synonyms for master and slave, respectively, in named.conf.

named will now log a warning if the old root DNSSEC key is explicitly configured and has not been updated. [RT #43670]

dig +nssearch will now list name servers that have timed out, in addition to those that respond. [GL #64]

Up to 64 response-policy zones are now supported by default; previously the limit was 32. [GL #123]

Several configuration options for time periods can now use TTL value suffixes (for example, 2h or 1d) in addition to an integer number of seconds. These include fstrm-set-reopen-interval, interface-interval, max-cache-ttl, max-ncache-ttl, max-policy-ttl, and min-update-interval. [GL #203]

NSID logging (enabled by the request-nsid option) now has its own nsid category, instead of using the resolver category.

The rndc nta command could not differentiate between views of the same name but different class; this has been corrected with the addition of a -class option. [GL #105]

allow-recursion-on and allow-query-cache-on each now default to the other if only one of them is set, in order to be consistent with the way allow-recursion and allow-query-cache work. [GL #319]

When compiled with IDN support, the dig and nslookup commands now disable IDN processing when the standard output is not a TTY (i.e., when the output is not being read by a human). When running from a shell script, the command line options +idnin and +idnout may be used to enable IDN processing of input and output domain names, respectively. When running on a TTY, the +noidnin and +noidnout options may be used to disable IDN processing of input and output domain names.

The configuration option max-ncache-ttl cannot exceed seven days. Previously, larger values than this were silently lowered; now, they trigger a configuration error.

The new dig -r command line option disables reading of the file $HOME/.digrc.

Zone signing and key maintenance events are now logged to the dnssec category rather than zone.