Release Notes

Introduction

BIND 9.17 is an unstable development release of BIND. This document summarizes new features and functional changes that have been introduced on this branch. With each development release leading up to the stable BIND 9.18 release, this document will be updated with additional features added and bugs fixed. Please see the CHANGES file for a more detailed list of changes and bug fixes.

Supported Platforms

To build on Unix-like systems, BIND requires support for POSIX.1c threads (IEEE Std 1003.1c-1995), the Advanced Sockets API for IPv6 (RFC 3542), and standard atomic operations provided by the C compiler.

The libuv asynchronous I/O library and the OpenSSL cryptography library must be available for the target platform. A PKCS#11 provider can be used instead of OpenSSL for Public Key cryptography (i.e., DNSSEC signing and validation), but OpenSSL is still required for general cryptography operations such as hashing and random number generation.

More information can be found in the PLATFORMS.md file that is included in the source distribution of BIND 9. If your compiler and system libraries provide the above features, BIND 9 should compile and run. If that is not the case, the BIND development team will generally accept patches that add support for systems that are still supported by their respective vendors.

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, source code, and pre-compiled versions for Microsoft Windows operating systems.

Notes for BIND 9.17.9

New Features

  • ipv4only.arpa is now served when DNS64 is configured. [GL #385]

Feature Changes

  • It is now possible to transition a zone from secure to insecure mode without making it bogus in the process; changing to dnssec-policy none; also causes CDS and CDNSKEY DELETE records to be published, to signal that the entire DS RRset at the parent must be removed, as described in RFC 8078. [GL #1750]

  • When using the unixtime or date method to update the SOA serial number, named and dnssec-signzone silently fell back to the increment method to prevent the new serial number from being smaller than the old serial number (using serial number arithmetics). dnssec-signzone now prints a warning message, and named logs a warning, when such a fallback happens. [GL #2058]

Bug Fixes

  • Multiple threads could attempt to destroy a single RBTDB instance at the same time, resulting in an unpredictable but low-probability assertion failure in free_rbtdb(). This has been fixed. [GL #2317]

  • named no longer attempts to assign threads to CPUs outside the CPU affinity set. Thanks to Ole Bjørn Hessen. [GL #2245]

  • When reconfiguring named, removing auto-dnssec did not turn off DNSSEC maintenance. This has been fixed. [GL #2341]

  • The report of intermittent BIND assertion failures triggered in lib/dns/resolver.c:dns_name_issubdomain() has now been closed without further action. Our initial response to this was to add diagnostic logging instead of terminating named, anticipating that we would receive further useful troubleshooting input. This workaround first appeared in BIND releases 9.17.5 and 9.16.7. However, since those releases were published, there have been no new reports of assertion failures matching this issue, but also no further diagnostic input, so we have closed the issue. [GL #2091]

Notes for BIND 9.17.8

New Features

  • NSEC3 support was added to KASP. A new option for dnssec-policy, nsec3param, can be used to set the desired NSEC3 parameters. NSEC3 salt collisions are automatically prevented during resalting. [GL #1620]

  • dig output now includes the transport protocol used (UDP, TCP, or TLS). [GL #1816]

  • dig can now report the DNS64 prefixes in use (+dns64prefix). This is useful when the host on which dig is run is behind an IPv6-only link, using DNS64/NAT64 or 464XLAT for IPv4aaS (IPv4 as a Service). [GL #1154]

Feature Changes

  • The new networking code introduced in BIND 9.16 (netmgr) was overhauled in order to make it more stable, testable, and maintainable. [GL #2321]

  • Earlier releases of BIND versions 9.16 and newer required the operating system to support load-balanced sockets in order for named to be able to achieve high performance (by distributing incoming queries among multiple threads). However, the only operating systems currently known to support load-balanced sockets are Linux and FreeBSD 12, which means both UDP and TCP performance were limited to a single thread on other systems. As of BIND 9.17.8, named attempts to distribute incoming queries among multiple threads on systems which lack support for load-balanced sockets (except Windows). [GL #2137]

  • The default value of max-recursion-queries was increased from 75 to 100. Since the queries sent towards root and TLD servers are now included in the count (as a result of the fix for CVE-2020-8616), max-recursion-queries has a higher chance of being exceeded by non-attack queries, which is the main reason for increasing its default value. [GL #2305]

  • The default value of nocookie-udp-size was restored back to 4096 bytes. Since max-udp-size is the upper bound for nocookie-udp-size, this change relieves the operator from having to change nocookie-udp-size together with max-udp-size in order to increase the default EDNS buffer size limit. nocookie-udp-size can still be set to a value lower than max-udp-size, if desired. [GL #2250]

Bug Fixes

  • Handling of missing DNS COOKIE responses over UDP was tightened by falling back to TCP. [GL #2275]

  • The CNAME synthesized from a DNAME was incorrectly followed when the QTYPE was CNAME or ANY. [GL #2280]

  • Building with native PKCS#11 support for AEP Keyper has been broken since BIND 9.17.4. This has been fixed. [GL #2315]

Notes for BIND 9.17.7

New Features

  • Support for DNS over TLS (DoT) has been added: the dig tool is now able to send DoT queries (+tls option) and named can handle DoT queries (listen-on tls ... option). named can use either a certificate provided by the user or an ephemeral certificate generated automatically upon startup. [GL #1840]

  • A new configuration option, stale-refresh-time, has been introduced. It allows a stale RRset to be served directly from cache for a period of time after a failed lookup, before a new attempt to refresh it is made. [GL #2066]

Feature Changes

  • The dig, host, and nslookup tools have been converted to use the new network manager API rather than the older ISC socket API.

    As a side effect of this change, the dig +unexpected option no longer works. This could previously be used to diagnose broken servers or network configurations by listening for replies from servers other than the one that was queried. With the new API, such answers are filtered before they ever reach dig, so the option has been removed. [GL #2140]

  • The network manager API is now used by named to send zone transfer requests. [GL #2016]

Bug Fixes

  • named could crash with an assertion failure if a TCP connection were closed while a request was still being processed. [GL #2227]

  • named acting as a resolver could incorrectly treat signed zones with no DS record at the parent as bogus. Such zones should be treated as insecure. This has been fixed. [GL #2236]

  • After a Negative Trust Anchor (NTA) is added, BIND performs periodic checks to see if it is still necessary. If BIND encountered a failure while creating a query to perform such a check, it attempted to dereference a NULL pointer, resulting in a crash. [GL #2244]

  • A problem obtaining glue records could prevent a stub zone from functioning properly, if the authoritative server for the zone were configured for minimal responses. [GL #1736]

  • UV_EOF is no longer treated as a TCP4RecvErr or a TCP6RecvErr. [GL #2208]

Notes for BIND 9.17.6

New Features

  • Add a new rndc command, rndc dnssec -rollover, which triggers a manual rollover for a specific key. [GL #1749]

  • Add a new rndc command, rndc dumpdb -expired, which dumps the cache database, including expired RRsets that are awaiting cleanup, to the dump-file for diagnostic purposes. [GL #1870]

Removed Features

  • The glue-cache option has been marked as deprecated. The glue cache feature still works and will be permanently enabled in a future release. [GL #2146]

Feature Changes

  • DNS Flag Day 2020: The default EDNS buffer size has been changed from 4096 to 1232 bytes, the EDNS buffer size probing has been removed, and named now sets the DF (Don’t Fragment) flag on outgoing UDP packets. According to measurements done by multiple parties, this should not cause any operational problems as most of the Internet “core” is able to cope with IP message sizes between 1400-1500 bytes; the 1232 size was picked as a conservative minimal number that could be changed by the DNS operator to an estimated path MTU minus the estimated header space. In practice, the smallest MTU witnessed in the operational DNS community is 1500 octets, the maximum Ethernet payload size, so a useful default for maximum DNS/UDP payload size on reliable networks would be 1400 bytes. [GL #2183]

Bug Fixes

  • named reported an invalid memory size when running in an environment that did not properly report the number of available memory pages and/or the size of each memory page. [GL #2166]

  • With multiple forwarders configured, named could fail the REQUIRE(msg->state == (-1)) assertion in lib/dns/message.c, causing it to crash. This has been fixed. [GL #2124]

  • named erroneously performed continuous key rollovers for KASP policies that used algorithm Ed25519 or Ed448 due to a mismatch between created key size and expected key size. [GL #2171]

  • Updating contents of an RPZ zone which contained names spelled using varying letter case could cause some processing rules in that RPZ zone to be erroneously ignored. [GL #2169]

Notes for BIND 9.17.5

New Features

  • Add a new rndc command, rndc dnssec -checkds, which signals to named that a DS record for a given zone or key has been published or withdrawn from the parent. This command replaces the time-based parent-registration-delay configuration option. [GL #1613]

  • Log when named adds a CDS/CDNSKEY to the zone. [GL #1748]

Removed Features

  • The --with-gperftools-profiler configure option was removed. To use the gperftools profiler, the HAVE_GPERFTOOLS_PROFILER macro now needs to be manually set in CFLAGS and -lprofiler needs to be present in LDFLAGS. [GL !4045]

  • The glue-cache option has been marked as deprecated. The glue cache feature still works and will be permanently enabled in a future release. [GL #2146]

Feature Changes

  • Previously, using dig +bufsize=0 had the side effect of disabling EDNS, and there was no way to test the remote server’s behavior when it had received a packet with EDNS0 buffer size set to 0. This is no longer the case; dig +bufsize=0 now sends a DNS message with EDNS version 0 and buffer size set to 0. To disable EDNS, use dig +noedns. [GL #2054]

Bug Fixes

  • In rare circumstances, named would exit with an assertion failure when the number of nodes stored in the red-black tree exceeded the maximum allowed size of the internal hash table. [GL #2104]

  • Silence spurious system log messages for an EPROTO(71) error code that was seen on older operating systems, where unhandled ICMPv6 errors resulted in a generic protocol error being returned instead of a more specific error code. [GL #1928]

  • With query name minimization enabled, named failed to resolve ip6.arpa. names that had extra labels to the left of the IPv6 part. For example, when named attempted query name minimization on a name like A.B.1.2.3.4.(...).ip6.arpa., it stopped at the leftmost IPv6 label, i.e. 1.2.3.4.(...).ip6.arpa., without considering the extra labels (A.B). That caused a query loop when resolving the name: if named received NXDOMAIN answers, then the same query was repeatedly sent until the number of queries sent reached the value of the max-recursion-queries configuration option. [GL #1847]

  • Parsing of LOC records was made more strict by rejecting a sole period (.) and/or m as a value. These changes prevent zone files using such values from being loaded. Handling of negative altitudes which are not integers was also corrected. [GL #2074]

  • Several problems found by OSS-Fuzz were fixed. (None of these are security issues.) [GL !3953] [GL !3975]

Notes for BIND 9.17.4

Security Fixes

  • It was possible to trigger an assertion failure by sending a specially crafted large TCP DNS message. This was disclosed in CVE-2020-8620.

    ISC would like to thank Emanuel Almeida of Cisco Systems, Inc. for bringing this vulnerability to our attention. [GL #1996]

  • named could crash after failing an assertion check in certain query resolution scenarios where QNAME minimization and forwarding were both enabled. To prevent such crashes, QNAME minimization is now always disabled for a given query resolution process, if forwarders are used at any point. This was disclosed in CVE-2020-8621.

    ISC would like to thank Joseph Gullo for bringing this vulnerability to our attention. [GL #1997]

  • It was possible to trigger an assertion failure when verifying the response to a TSIG-signed request. This was disclosed in CVE-2020-8622.

    ISC would like to thank Dave Feldman, Jeff Warren, and Joel Cunningham of Oracle for bringing this vulnerability to our attention. [GL #2028]

  • When BIND 9 was compiled with native PKCS#11 support, it was possible to trigger an assertion failure in code determining the number of bits in the PKCS#11 RSA public key with a specially crafted packet. This was disclosed in CVE-2020-8623.

    ISC would like to thank Lyu Chiy for bringing this vulnerability to our attention. [GL #2037]

  • update-policy rules of type subdomain were incorrectly treated as zonesub rules, which allowed keys used in subdomain rules to update names outside of the specified subdomains. The problem was fixed by making sure subdomain rules are again processed as described in the ARM. This was disclosed in CVE-2020-8624.

    ISC would like to thank Joop Boonen of credativ GmbH for bringing this vulnerability to our attention. [GL #2055]

New Features

  • A new configuration option stale-cache-enable has been introduced to enable or disable keeping stale answers in cache. [GL #1712]

  • rndc has been updated to use the new BIND network manager API. This change had the side effect of altering the TCP timeout for RNDC connections from 60 seconds to the tcp-idle-timeout value, which defaults to 30 seconds. Also, because the network manager currently has no support for UNIX-domain sockets, those cannot now be used with rndc. This will be addressed in a future release, either by restoring UNIX-domain socket support or by formally declaring them to be obsolete in the control channel. [GL #1759]

  • Statistics channels have also been updated to use the new BIND network manager API. [GL #2022]

Feature Changes

  • BIND’s cache database implementation has been updated to use a faster hash function with better distribution. In addition, the effective max-cache-size (configured explicitly, defaulting to a value based on system memory or set to unlimited) now pre-allocates fixed-size hash tables. This prevents interruption to query resolution when the hash table sizes need to be increased. [GL #1775]

  • Keeping stale answers in cache has been disabled by default. [GL #1712]

  • Resource records received with 0 TTL are no longer kept in the cache to be used for stale answers. [GL #1829]

Bug Fixes

  • Wildcard RPZ passthru rules could incorrectly be overridden by other rules that were loaded from RPZ zones which appeared later in the response-policy statement. This has been fixed. [GL #1619]

  • The IPv6 Duplicate Address Detection (DAD) mechanism could inadvertently prevent named from binding to new IPv6 interfaces, by causing multiple route socket messages to be sent for each IPv6 address. named monitors for new interfaces to bind() to when it is configured to listen on any or on a specific range of addresses. New IPv6 interfaces can be in a “tentative” state before they are fully available for use. When DAD is in use, two messages are emitted by the route socket: one when the interface first appears and then a second one when it is fully “up.” An attempt by named to bind() to the new interface prematurely would fail, causing it thereafter to ignore that address/interface. The problem was worked around by setting the IP_FREEBIND option on the socket and trying to bind() to each IPv6 address again if the first bind() call for that address failed with EADDRNOTAVAIL. [GL #2038]

  • Addressed an error in recursive clients stats reporting which could cause underflow, and even negative statistics. There were occasions when an incoming query could trigger a prefetch for some eligible RRset, and if the prefetch code were executed before recursion, no increment in recursive clients stats would take place. Conversely, when processing the answers, if the recursion code were executed before the prefetch, the same counter would be decremented without a matching increment. [GL #1719]

  • The introduction of KASP support inadvertently caused the second field of sig-validity-interval to always be calculated in hours, even in cases when it should have been calculated in days. This has been fixed. (Thanks to Tony Finch.) [GL !3735]

  • LMDB locking code was revised to make rndc reconfig work properly on FreeBSD and with LMDB >= 0.9.26. [GL #1976]

Notes for BIND 9.17.3

New Features

  • New rndc command rndc dnssec -status shows the current DNSSEC policy and keys in use, the key states, and rollover status. [GL #1612]

  • Added support in the network manager for initiating outgoing TCP connections. [GL #1958]

Feature Changes

  • Disable and disallow static linking of BIND 9 binaries and libraries as BIND 9 modules require dlopen() support and static linking also prevents using security features like read-only relocations (RELRO) or address space layout randomization (ASLR) which are important for programs that interact with the network and process arbitrary user input. [GL #1933]

  • As part of an ongoing effort to use RFC 8499 terminology, primaries can now be used as a synonym for masters in named.conf. Similarly, notify primary-only can now be used as a synonym for notify master-only. The output of rndc zonestatus now uses primary and secondary terminology. [GL #1948]

Bug Fixes

  • A race condition could occur if a TCP socket connection was closed while named was waiting for a recursive response. The attempt to send a response over the closing connection triggered an assertion failure in the function isc__nm_tcpdns_send(). [GL #1937]

  • A race condition could occur when named attempted to use a UDP interface that was shutting down. This triggered an assertion failure in uv__udp_finish_close(). [GL #1938]

  • Fix assertion failure when server was under load and root zone had not yet been loaded. [GL #1862]

  • named could crash when cleaning dead nodes in lib/dns/rbtdb.c that were being reused. [GL #1968]

  • named crashed on shutdown when a new rndc connection was received during shutdown. This has been fixed. [GL #1747]

  • The DS RRset returned by dns_keynode_dsset() was used in a non-thread-safe manner. This could result in an INSIST being triggered. [GL #1926]

  • The primary and secondary keywords, when used as parameters for check-names, were not processed correctly and were being ignored. [GL #1949]

  • rndc dnstap -roll <value> did not limit the number of saved files to <value>. [GL !3728]

  • The validator could fail to accept a properly signed RRset if an unsupported algorithm appeared earlier in the DNSKEY RRset than a supported algorithm. It could also stop if it detected a malformed public key. [GL #1689]

  • The blackhole ACL was inadvertently disabled for client queries. Blocked IP addresses were not used for upstream queries but queries from those addresses could still be answered. [GL #1936]

Notes for BIND 9.17.2

Security Fixes

  • To prevent exhaustion of server resources by a maliciously configured domain, the number of recursive queries that can be triggered by a request before aborting recursion has been further limited. Root and top-level domain servers are no longer exempt from the max-recursion-queries limit. Fetches for missing name server address records are limited to 4 for any domain. This issue was disclosed in CVE-2020-8616. [GL #1388]

  • Replaying a TSIG BADTIME response as a request could trigger an assertion failure. This was disclosed in CVE-2020-8617. [GL #1703]

  • It was possible to trigger an assertion when attempting to fill an oversized TCP buffer. This was disclosed in CVE-2020-8618. [GL #1850]

  • It was possible to trigger an INSIST failure when a zone with an interior wildcard label was queried in a certain pattern. This was disclosed in CVE-2020-8619. [GL #1111] [GL #1718]

Known Issues

  • In this release, the build system has been significantly changed (see below) and there are several unresolved issues to be aware of when using a development release. Please refer to GitLab issue #4 for a list of not-yet-resolved issues that will be fixed in future releases. [GL #4]

  • BIND crashes on startup when linked against libuv 1.36. This issue is related to recvmmsg() support in libuv, which was first included in libuv 1.35. The problem was addressed in libuv 1.37, but the relevant libuv code change requires a special flag to be set during library initialization in order for recvmmsg() support to be enabled. This BIND release sets that special flag when required, so recvmmsg() support is now enabled when BIND is compiled against either libuv 1.35 or libuv 1.37+; libuv 1.36 is still not usable with BIND. [GL #1761] [GL #1797]

New Features

  • The BIND 9 build system has been changed to use a typical autoconf+automake+libtool stack. This should not make any difference for people building BIND 9 from release tarballs, but when building BIND 9 from the Git repository, autoreconf -fi needs to be run first. Extra attention is also needed when using non-standard ./configure options. [GL #4]

  • Documentation was converted from DocBook to reStructuredText. The BIND 9 ARM is now generated using Sphinx and published on Read the Docs. Release notes are no longer available as a separate document accompanying a release. [GL #83]

  • named and named-checkzone now reject master zones that have a DS RRset at the zone apex. Attempts to add DS records at the zone apex via UPDATE will be logged but otherwise ignored. DS records belong in the parent zone, not at the zone apex. [GL #1798]

  • Per-type record count limits can now be specified in update-policy statements, to limit the number of records of a particular type that can be added to a domain name via dynamic update. [GL #1657]

  • dig and other tools can now print the Extended DNS Error (EDE) option when it appears in a request or a response. [GL #1835]

  • dig +qid=<num> allows the user to specify a particular query ID for testing purposes. [GL #1851]

  • A new logging category, rpz-passthru, was added, which allows RPZ passthru actions to be logged into a separate channel. [GL #54]

  • Zone timers are now exported via statistics channel. For primary zones, only the load time is exported. For secondary zones, exported timers also include expire and refresh times. Contributed by Paul Frieden, Verizon Media. [GL #1232]

Feature Changes

  • The default value of max-stale-ttl has changed from 1 week to 12 hours. This option controls how long named retains expired RRsets in cache as a potential mitigation mechanism, should there be a problem with one or more domains. Note that cache content retention is independent of whether stale answers are used in response to client queries (stale-answer-enable yes|no and rndc serve-stale on|off). Serving of stale answers when the authoritative servers are not responding must be explicitly enabled, whereas the retention of expired cache content takes place automatically on all versions of BIND 9 that have this feature available. [GL #1877]

    Warning

    This change may be significant for administrators who expect that stale cache content will be automatically retained for up to 1 week. Add option max-stale-ttl 1w; to named.conf to keep the previous behavior of named.

  • BIND 9 no longer sets receive/send buffer sizes for UDP sockets, relying on system defaults instead. [GL #1713]

  • The default rwlock implementation has been changed back to the native BIND 9 rwlock implementation. [GL #1753]

  • BIND 9 binaries which are neither daemons nor administrative programs were moved to $bindir. Only ddns-confgen, named, rndc, rndc-confgen, and tsig-confgen were left in $sbindir. [GL #1724]

  • listen-on-v6 { any; } creates a separate socket for each interface. Previously, just one socket was created on systems conforming to RFC 3493 and RFC 3542. This change was introduced in BIND 9.16.0, but it was accidentally omitted from documentation. [GL #1782]

  • The native PKCS#11 EdDSA implementation has been updated to PKCS#11 v3.0 and thus made operational again. Contributed by Aaron Thompson. [GL !3326]

  • The OpenSSL ECDSA implementation has been updated to support PKCS#11 via OpenSSL engine (see engine_pkcs11 from libp11 project). [GL #1534]

  • The OpenSSL EdDSA implementation has been updated to support PKCS#11 via OpenSSL engine. Please note that an EdDSA-capable OpenSSL engine is required and thus this code is only a proof-of-concept for the time being. Contributed by Aaron Thompson. [GL #1763]

  • Message IDs in inbound AXFR transfers are now checked for consistency. Log messages are emitted for streams with inconsistent message IDs. [GL #1674]

  • The question section is now checked when processing AXFR, IXFR, and SOA replies while transferring a zone in. [GL #1683]

Bug Fixes

  • When fully updating the NSEC3 chain for a large zone via IXFR, a temporary loss of performance could be experienced on the secondary server when answering queries for nonexistent data that required DNSSEC proof of non-existence (in other words, queries that required the server to find and to return NSEC3 data). The unnecessary processing step that was causing this delay has now been removed. [GL #1834]

  • named could crash with an assertion failure if the name of a database node was looked up while the database was being modified. [GL #1857]

  • When running on a system with support for Linux capabilities, named drops root privileges very soon after system startup. This was causing a spurious log message, unable to set effective uid to 0: Operation not permitted, which has now been silenced. [GL #1042] [GL #1090]

  • A possible deadlock in lib/isc/unix/socket.c was fixed. [GL #1859]

  • Previously, named did not destroy some mutexes and conditional variables in netmgr code, which caused a memory leak on FreeBSD. This has been fixed. [GL #1893]

  • A data race in lib/dns/resolver.c:log_formerr() that could lead to an assertion failure was fixed. [GL #1808]

  • Previously, provide-ixfr no; failed to return up-to-date responses when the serial number was greater than or equal to the current serial number. [GL #1714]

  • A bug in dnstap initialization could prevent some dnstap data from being logged, especially on recursive resolvers. [GL #1795]

  • A bug in dnssec-policy keymgr was fixed, where the check for the existence of a given key’s successor would incorrectly return true if any other key in the keyring had a successor. [GL #1845]

  • With dnssec-policy, when creating a successor key, the “goal” state of the current active key (the predecessor) was not changed and thus never removed from the zone. [GL #1846]

  • When named-checkconf -z was run, it would sometimes incorrectly set its exit code. It reflected the status of the last view found; if zone-loading errors were found in earlier configured views but not in the last one, the exit code indicated success. Thanks to Graham Clinch. [GL #1807]

  • named-checkconf -p could include spurious text in server-addresses statements due to an uninitialized DSCP value. This has been fixed. [GL #1812]

  • When built without LMDB support, named failed to restart after a zone with a double quote (“) in its name was added with rndc addzone. Thanks to Alberto Fernández. [GL #1695]

  • The ARM has been updated to indicate that the TSIG session key is generated when named starts, regardless of whether it is needed. [GL #1842]

Notes for BIND 9.17.1

Security Fixes

  • DNS rebinding protection was ineffective when BIND 9 is configured as a forwarding DNS server. Found and responsibly reported by Tobias Klein. [GL #1574]

Known Issues

  • We have received reports that in some circumstances, receipt of an IXFR can cause the processing of queries to slow significantly. Some of these were related to RPZ processing, which has been fixed in this release (see below). Others appear to occur where there are NSEC3-related changes (such as an operator changing the NSEC3 salt used in the hash calculation). These are being investigated. [GL #1685]

New Features

  • A new option, nsdname-wait-recurse, has been added to the response-policy clause in the configuration file. When set to no, RPZ NSDNAME rules are only applied if the authoritative nameservers for the query name have been looked up and are present in the cache. If this information is not present, the RPZ NSDNAME rules are ignored, but the information is looked up in the background and applied to subsequent queries. The default is yes, meaning that RPZ NSDNAME rules should always be applied, even if the information needs to be looked up first. [GL #1138]

Feature Changes

  • The previous DNSSEC sign statistics used lots of memory. The number of keys to track is reduced to four per zone, which should be enough for 99% of all signed zones. [GL #1179]

Bug Fixes

  • When an RPZ policy zone was updated via zone transfer and a large number of records was deleted, named could become nonresponsive for a short period while deleted names were removed from the RPZ summary database. This database cleanup is now done incrementally over a longer period of time, reducing such delays. [GL #1447]

  • When trying to migrate an already-signed zone from auto-dnssec maintain to one based on dnssec-policy, the existing keys were immediately deleted and replaced with new ones. As the key rollover timing constraints were not being followed, it was possible that some clients would not have been able to validate responses until all old DNSSEC information had timed out from caches. BIND now looks at the time metadata of the existing keys and incorporates it into its DNSSEC policy operation. [GL #1706]

Notes for BIND 9.17.0

Known Issues

  • UDP network ports used for listening can no longer simultaneously be used for sending traffic. An example configuration which triggers this issue would be one which uses the same address:port pair for listen-on(-v6) statements as for notify-source(-v6) or transfer-source(-v6). While this issue affects all operating systems, it only triggers log messages (e.g. “unable to create dispatch for reserved port”) on some of them. There are currently no plans to make such a combination of settings work again.

New Features

  • When a secondary server receives a large incremental zone transfer (IXFR), it can have a negative impact on query performance while the incremental changes are applied to the zone. To address this, named can now limit the size of IXFR responses it sends in response to zone transfer requests. If an IXFR response would be larger than an AXFR of the entire zone, it will send an AXFR response instead.

    This behavior is controlled by the max-ixfr-ratio option - a percentage value representing the ratio of IXFR size to the size of a full zone transfer. The default is 100%. [GL #1515]

  • A new RPZ option nsdname-wait-recurse controls whether RPZ-NSDNAME rules should always be applied even if the names of authoritative name servers for the query name need to be looked up recurively first. The default is yes. Setting it to no speeds up initial responses by skipping RPZ-NSDNAME rules when name server domain names are not yet in the cache. The names will be looked up in the background and the rule will be applied for subsequent queries. [GL #1138]

Feature Changes

  • The system-provided POSIX Threads read-write lock implementation is now used by default instead of the native BIND 9 implementation. Please be aware that glibc versions 2.26 through 2.29 had a bug that could cause BIND 9 to deadlock. A fix was released in glibc 2.30, and most current Linux distributions have patched or updated glibc, with the notable exception of Ubuntu 18.04 (Bionic) which is a work in progress. If you are running on an affected operating system, compile BIND 9 with --disable-pthread-rwlock until a fixed version of glibc is available. [GL !3125]

  • The rndc nta -dump and rndc secroots commands now both include validate-except entries when listing negative trust anchors. These are indicated by the keyword permanent in place of the expiry date. [GL #1532]

Bug Fixes

  • Fixed re-signing issues with inline zones which resulted in records being re-signed late or not at all.

License

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

The license requires that if you make changes to BIND and distribute them outside your organization, those changes must be published under the same license. It does not require that you publish or disclose anything other than the changes you have made to our software. This requirement does not affect anyone who is using BIND, with or without modifications, without redistributing it, nor anyone redistributing BIND without changes.

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

End of Life

BIND 9.17 is an unstable development branch. When its development is complete, it will be renamed to BIND 9.18, which will be a stable branch. The end-of-life date for BIND 9.18 has not yet been determined. For those needing long-term stability, the current Extended Support Version (ESV) is BIND 9.11, which will be supported until at least December 2021. 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.