Michal Nowak [Mon, 9 Sep 2019 07:35:02 +0000 (09:35 +0200)]
Various README.md and README fixes
Fixing typos, typographical glitches. Added backticks around binaries,
modules, and libraries so it's more consistent. Added a paragraph with
ISC Security Policy.
Michał Kępień [Thu, 12 Sep 2019 12:25:57 +0000 (14:25 +0200)]
Add FreeBSD to GitLab CI
Ensure BIND can be tested on FreeBSD in GitLab to more quickly catch
build and test errors on that operating system. Make the relevant jobs
optional until the CI environment supporting them is deemed stable
enough for continuous use.
FreeBSD jobs are run using the Custom executor feature of GitLab Runner.
Unlike the Docker executor, the Custom executor does not support the
"image" option and thus some way of informing the runner about the OS
version to use for a given job is necessary. Arguably the simplest way
of doing that without a lot of code duplication in .gitlab-ci.yml would
be to use a YAML template with a "variables" block specifying the
desired FreeBSD release to use, but including such a template in a job
definition would cause issues in case other variables also needed to be
set for that job (e.g. CFLAGS or EXTRA_CONFIGURE for build jobs). Thus,
only one FreeBSD YAML template is defined instead and the Custom
executor scripts on FreeBSD runners extract the OS version to use from
the CI job name. This allows .gitlab-ci.yml variables to be defined for
FreeBSD jobs in the same way as for Docker-based jobs.
Michał Kępień [Thu, 12 Sep 2019 12:25:57 +0000 (14:25 +0200)]
Tweak lib/dns/tests/tkey_test requirements
Currently, the lib/dns/tests/tkey_test unit test is only run when the
linker supports the --wrap option. However, linker support for that
option is only needed for static builds. As a result, the unit test
mentioned before is not being run everywhere it can be run as even for
builds done using --with-libtool, the test is not run unless the linker
supports the --wrap option.
Tweak preprocessor directives in lib/dns/tests/tkey_test.c so that this
test is run:
- for all builds using --with-libtool,
- for static builds done using a linker supporting the --wrap option.
Michał Kępień [Thu, 12 Sep 2019 12:25:57 +0000 (14:25 +0200)]
Make lib/dns/tests/tkey_test.c more portable
Weak symbols are handled differently by different dynamic linkers. With
glibc, lib/dns/tests/tkey_test works as expected no matter whether
--with-libtool is used or not: __attribute__((weak)) prevents a static
build from failing and it just so happens that the desired symbols are
picked at runtime for dynamic builds. However, with BSD libc, the
libdns functions called from lib/dns/tests/tkey_test.c use the "real"
memory allocation functions from libisc, thus breaking that unit test.
(Note: similar behavior can be reproduced with glibc by setting the
LD_DYNAMIC_WEAK environment variable.)
The simplest way to make lib/dns/tests/tkey_test work reliably is to
drop all uses of __attribute__((weak)) in it - this way, the memory
functions inside lib/dns/tests/tkey_test.c will always be used instead
of the "real" libisc ones for dynamic builds. However, this would not
work with static builds as it would result in multiple strong symbols
with the same name being present in a single binary.
Work around the problem by only compiling in the overriding definitions
of memory functions when building using --with-libtool. For static
builds, keep relying on the --wrap linker option for replacing calls to
the functions we are interested in.
Michał Kępień [Thu, 12 Sep 2019 12:25:57 +0000 (14:25 +0200)]
Set --logfile for all kyua invocations
When kyua is called without the --logfile command line option, the log
file is created at a default location which is derived from the HOME
environment variable. On FreeBSD GitLab CI runners, /home is a
read-only directory and thus kyua invocations not using the --logfile
option fail when HOME is set to something beneath /home. Set --logfile
to /dev/null for all kyua invocations whose logs are irrelevant in order
to prevent kyua failures caused by HOME being non-writable.
Michał Kępień [Thu, 12 Sep 2019 11:57:24 +0000 (13:57 +0200)]
README: do not mention /usr/include on macOS
For newer versions of Xcode, "xcode-select --install" no longer installs
system headers into /usr/include (instead, they are installed in the
Xcode directory tree), so do not mention that path in the macOS section
of README to prevent confusion.
Previously the libisc allocator had ability to run unlocked when threading was
disabled. As the threading is now always on, remove the ISC_MEMFLAG_NOLOCK
memory flag as it serves no purpose.
The isc_mem_createx() function was only used in the tests to eliminate using the
default flags (which as of writing this commit message was ISC_MEMFLAG_INTERNAL
and ISC_MEMFLAG_FILL). This commit removes the isc_mem_createx() function from
the public API.
Simplify isc_mem_create() to always use defaults and never fail
Previously, the isc_mem_create() and isc_mem_createx() functions took `max_size`
and `target_size` as first two arguments. Those values were never used in the
BIND 9 code. The refactoring removes those arguments and let BIND 9 always use
the default values.
Previously, the isc_mem_create() and isc_mem_createx() functions could have
failed because of failed memory allocation. As this was no longer true and the
functions have always returned ISC_R_SUCCESS, the have been refactored to return
void.
Ondřej Surý [Thu, 29 Aug 2019 11:59:52 +0000 (13:59 +0200)]
Add OpenSSL based isc_siphash24() implementation
This commits adds an OpenSSL based isc_siphash24() implementation, which is
preferred when available.
The siphash_test has been modified to test both implementation with a trick that
renames the isc_siphash24() to openssl_ or native_ prefixed name and includes
the ../siphash.c two times (when the OpenSSL implementation is available).
Ondřej Surý [Fri, 30 Aug 2019 13:32:11 +0000 (15:32 +0200)]
configure.ac: Add OpenSSL SipHash support detection
Add check for creating new EVP_PKEY with EVP_PKEY_SIPHASH, but disable SipHash
on OpenSSL 1.1.1 as the hash length initialization is broken before OpenSSL
1.1.1a release.
Ondřej Surý [Thu, 29 Aug 2019 11:59:52 +0000 (13:59 +0200)]
Fix alignment issues in the native implementation of isc_siphash24()
The native implementation's conversion from the uint8_t buffers to uint64_t now
follows the reference implementation that doesn't require aligned buffers.
Ondřej Surý [Thu, 29 Aug 2019 08:29:53 +0000 (10:29 +0200)]
isc_event_allocate() cannot fail, remove the fail handling blocks
isc_event_allocate() calls isc_mem_get() to allocate the event structure. As
isc_mem_get() cannot fail softly (e.g. it never returns NULL), the
isc_event_allocate() cannot return NULL, hence we remove the (ret == NULL)
handling blocks using the semantic patch from the previous commit.
Evan Hunt [Wed, 28 Aug 2019 20:11:35 +0000 (13:11 -0700)]
use an rbtnodechain to walk up labels
when looking for a possible wildcard match in the RPZ summary database,
use an rbtnodechain to walk up label by label, rather than using the
node's parent pointer.
Ondřej Surý [Thu, 29 Aug 2019 12:27:19 +0000 (14:27 +0200)]
Make use of DAG for GitLab Pipelines
GitLab 12.2 has introduced Directed Acyclic Graphs in the GitLab CI[1] that
allow jobs to run out-of-order and not wait for the whole previous stage to
complete.
projects / thirdparty / bind9.git / log
