[Unbound-users] Release of unbound 1.2.0
wouter at NLnetLabs.nl
Wed Jan 14 08:10:03 UTC 2009
-----BEGIN PGP SIGNED MESSAGE-----
Unbound 1.2.0 is released.
It has a long list of changes. The new featureset is small, but there
are important, security related, fixes.
Maintainers, bug#228 is fixed in this release, as well as an iana
portlist update, compared to the 1.2.0rc1 sent out to package
maintainers last week.
* Wildcard support for trusted-keys-file: "/etc/keys/*.key"
* unbound-control status command.
* extended statistics has a number of ipv6 queries counter.
contrib/unbound_munin_ was updated to draw ipv6 in the hits graph.
* SElinux policy files in contrib/selinux for the unbound daemon, by
Paul Wouters and Adam Tkac.
* The long standing bug with libevent use is fixed. It turns out to be a
race condition in the calls to libevent that only causes harm when very
busy. The builtin mini-event did not have a problem being called like
this, but libevent and libev usage is now fixed. Libevent 1.1 is
reported to still give problems, but 1.4.5 and 1.4.8 seem fine.
* Certain packets could cause an assertion failure. Resulting in a
denial-of-service vector if the server was compiled with --enable-debug
(assertions enabled). This is fixed.
* fixed bug reported by Duane Wessels: error in DLV lookup, would make
some zones that had correct DLV keys as insecure.
* [bugzilla: 228 ]
fix lame marking. security fix that resolves denial of service that
could be triggered by an unusual configuration. Thanks to Mark Zealey
Other Bug Fixes
* [bugzilla: 224 ]
no more race condition in makefile during built with high -j inside
included libldns version.
* iana portlist updated to most recent, avoids allocated ports.
* L root server AAAA record added to builtin root hints.
* removed possible race condition in unit test for race conditions.
* fixup reported problem with transparent local-zone data where queries
with different type could get nxdomain. Now queries with a different
name get resolved normally, with different type get a correct
* HINFO no longer downcased for validation, making unbound compatible
with bind and ldns.
* fix reading included config files when chrooted. Give full path names
for include files. Relative path names work if the start dir equals the
* fix libunbound message transport when no packet buffer is available.
* fixup getaddrinfo failure handling for remote control port.
* fixup so it works with libev-3.51 from http://dist.schmorp.de/libev/
* ldns tarball updated with 1.4.1rc for DLV unit test.
* fixup BSD port for infra host storage. It hashed wrongly.
* follow ldns rc makedist name generation.
* snapshot version uses _ not - to help rpm distinguish the version number.
* do not reopen syslog to avoid dev/log dependency. This makes chroot
* [bugzilla: 219 ]
better fix for bug #219: use LOG_NDELAY with openlog() call. Thanks to
* [bugzilla: 221 ]
fixed: unbound checkconf checks if key files exist if remote control is
enabled. Also fixed NULL printf when not chrooted.
* Fix problem reported by Jaco Engelbrecht where unbound-control stats
freezes up unbound if this was compiled without threading, and was using
* test for remote control with interprocess communication.
* created command distribution mechanism so that remote control commands
other than 'stats' work on all processes in a nonthreaded compiled
version. dump/load cache work, on the first process.
* fixup remote control local_data addition memory corruption bug.
* [bugzilla: 220 ]
configure complains when --without-ssl is given, fixed.
* blacklisted servers are polled at a low rate (1%) to see if they come
back up. But not if there is some other working server.
* documented that the user of the server daemon needs read privileges on
the keys and certificates generated by unbound-control-setup. This is
different per system or distribution, usually, running the script under
the same username as the server uses suffices. i.e. sudo -u unbound
* unbound-control-setup.sh removes read/write permissions other from the
keys it creates (as suggested by Dmitriy Demidov).
* fixed tcp accept, errors were printed when they should not.
* fixup fatal error due to faulty error checking after tcp accept.
* add check in rlimit code to avoid integer underflow.
* rlimit check with new formula; better estimate for number interfaces.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org
-----END PGP SIGNATURE-----
More information about the Unbound-users