D-Bus 1.8.6 (2014-06-02)
==
Security fixes:
• On Linux ≥ 2.6.37-rc4, if sendmsg() fails with ETOOMANYREFS, silently drop
the message. This prevents an attack in which a malicious client can
make dbus-daemon disconnect a system service, which is a local
denial of service.
(fd.o #80163, CVE-2014-3532; Alban Crequy)
• Track remaining Unix file descriptors correctly when more than one
message in quick succession contains fds. This prevents another attack
in which a malicious client can make dbus-daemon disconnect a system
service.
(fd.o #79694, fd.o #80469, CVE-2014-3533; Alejandro Martínez Suárez,
Simon McVittie, Alban Crequy)
Other fixes:
• When dbus-launch --exit-with-session starts a dbus-daemon but then cannot
attach to a session, kill the dbus-daemon as intended
(fd.o #74698, Роман Донченко)
D-Bus 1.8.4 (2014-06-10)
==
Security fix:
• Alban Crequy at Collabora Ltd. discovered and fixed a denial-of-service
flaw in dbus-daemon, part of the reference implementation of D-Bus.
Additionally, in highly unusual environments the same flaw could lead to
a side channel between processes that should not be able to communicate.
(CVE-2014-3477, fd.o #78979)
D-Bus 1.8.2 (2014-04-30)
==
The “nobody wants red” release.
Enhancements:
• in the CMake build system, add some hints for Linux users cross-compiling
Windows D-Bus binaries to be able to run tests under Wine
(fd.o #41252, Ralf Habacker)
• add Documentation key to dbus.service (fd.o #77447, Cameron Norman)
Fixes:
• in "dbus-uuidgen --ensure", try to copy systemd's /etc/machine-id
to /var/lib/dbus/machine-id instead of generating an entirely new ID
(fd.o #77941, Simon McVittie)
• if dbus-launch receives an X error very quickly, do not kill
unrelated processes (fd.o #74698, Роман Донченко)
• on Windows, allow up to 8K connections to the dbus-daemon, instead of the
previous 64 (fd.o #71297; Cristian Onet, Ralf Habacker)
• cope with \r\n newlines in regression tests, since on Windows,
dbus-daemon.exe uses text mode (fd.o #75863, Руслан Ижбулатов)
D-Bus 1.8.0 (2014-01-20)
==
The “Wolverine distrusts my printer” release.
This starts a new stable branch. The 1.6.x branch is now considered to be
outdated, and will only receive fixes for serious bugs such as security
flaws. The 1.4.x and 1.2.x branches no longer have upstream support and
are unlikely to get any more releases, but if distributors still need to
support them, please share security patches via upstream.
Summary of changes since 1.6.x:
• libdbus always behaves as if dbus_threads_init_default() had been called
(thread-safety by default)
• new dbus-run-session tool, replacing certain misuses of dbus-launch
• dbus-monitor can talk to outdated versions of dbus-daemon again
• new org.freedesktop.DBus.GetConnectionCredentials method
• GetConnectionUnixProcessID also works correctly on Windows, returning
the Windows process ID
• GetConnectionWindowsSID returns the correct SID on Windows
• expat is required, libxml2 can no longer be used as a substitute
• the userDB cache is required, and cannot be disabled
• a 64-bit integer type (either int, long, long long or _int64) is required
• better systemd-journald integration on Linux
• fixed long-standing fd and array leaks when failing to parse a message
• fixed referenced-but-never-freed parent nodes (effectively memory leaks)
when using certain object-path allocation patterns, notably in Avahi
• better defaults for Windows support
• better CMake support
• better portability to mingw32, FreeBSD, NetBSD, QNX and Hurd
• the source language for the man pages is now Docbook XML
Enhancements since 1.7.10:
• Enhance the CMake build system to check for GLib and compile/run
a subset of the regression tests (fd.o #41252, #73495; Ralf Habacker)
Fixes since 1.7.10:
• don't rely on va_copy(), use DBUS_VA_COPY() wrapper (fd.o #72840,
Ralf Habacker)
• fix compilation of systemd journal support on older systemd versions where
sd-journal.h doesn't include syslog.h (fd.o #73455, Ralf Habacker)
• fix compilation on older MSVC versions by including stdlib.h
(fd.o #73455, Ralf Habacker)
• Allow <allow_anonymous/> to appear in an included configuration file
(fd.o #73475, Matt Hoosier)
Test behaviour changes since 1.7.10:
• If the tests crash with an assertion failure, they no longer default to
blocking for a debugger to be attached. Set DBUS_BLOCK_ON_ABORT in the
environment if you want the old behaviour.
• To improve debuggability, the dbus-daemon and dbus-daemon-eavesdrop tests
can be run with an external dbus-daemon by setting
DBUS_TEST_DAEMON_ADDRESS in the environment. Test-cases that require
an unusually-configured dbus-daemon are skipped.
D-Bus 1.7.10 (2014-01-06)
==
The “weighted companion cube” release.
This is a release candidate for D-Bus 1.8.
D-Bus Specification 0.23:
• don't require messages with no INTERFACE to be dispatched
(fd.o #68597, Simon McVittie)
• document "tcp:bind=..." and "nonce-tcp:bind=..." (fd.o #72301,
Chengwei Yang)
• define "listenable" and "connectable" addresses, and discuss
the difference (fd.o #61303, Simon McVittie)
Enhancements:
• support printing Unix file descriptors in dbus-send, dbus-monitor
(fd.o #70592, Robert Ancell)
• don't install systemd units if --disable-systemd is given
(fd.o #71818, Chengwei Yang)
Fixes:
• don't leak memory on out-of-memory while listing activatable or
active services (fd.o #71526, Radoslaw Pajak)
• fix undefined behaviour in a regression test (fd.o #69924, DreamNik)
• escape Unix socket addresses correctly (fd.o #46013, Chengwei Yang)
• on SELinux systems, don't assume that SECCLASS_DBUS, DBUS__ACQUIRE_SVC
and DBUS__SEND_MSG are numerically equal to their values in the
reference policy (fd.o #88719, osmond sun)
• define PROCESS_QUERY_LIMITED_INFORMATION if missing from MinGW < 4 headers
(fd.o #71366, Matt Fischer)
• define WIN32_LEAN_AND_MEAN to avoid conflicts between winsock.h and
winsock2.h (fd.o #71405, Matt Fischer)
• do not return failure from _dbus_read_nonce() with no error set,
preventing a potential crash (fd.o #72298, Chengwei Yang)
• on BSD systems, avoid some O(1)-per-process memory and fd leaks in kqueue,
preventing test failures (fd.o #69332, fd.o #72213; Chengwei Yang)
• fix warning spam on Hurd by not trying to set SO_REUSEADDR on Unix sockets,
which doesn't do anything anyway on at least Linux and FreeBSD
(fd.o #69492, Simon McVittie)
• fix use of TCP sockets on FreeBSD and Hurd by tolerating EINVAL from
sendmsg() with SCM_CREDS (retrying with plain send()), and looking
for credentials more correctly (fd.o #69492, Simon McVittie)
• ensure that tests run with a temporary XDG_RUNTIME_DIR to avoid
getting mixed up in XDG/systemd "user sessions" (fd.o #61301,
Simon McVittie)
• refresh cached policy rules for existing connections when bus
configuration changes (fd.o #39463, Chengwei Yang)
D-Bus 1.7.8 (2013-11-01)
==
The “extreme hills” release.
Dependencies:
• If systemd support is enabled, libsystemd-journal is now required.
Enhancements:
• When activating a non-systemd service under systemd, annotate its
stdout/stderr with its bus name in the Journal. Known limitation:
because the socket is opened before forking, the process will still be
logged as if it had dbus-daemon's process ID and user ID.
(fd.o #68559, Chengwei Yang)
• Document more configuration elements in dbus-daemon(1)
(fd.o #69125, Chengwei Yang)
Fixes:
• Don't leak string arrays or fds if dbus_message_iter_get_args_valist()
unpacks them and then encounters an error (fd.o #21259, Chengwei Yang)
• If compiled with libaudit, retain CAP_AUDIT_WRITE so we can write
disallowed method calls to the audit log, fixing a regression in 1.7.6
(fd.o #49062, Colin Walters)
• path_namespace='/' in match rules incorrectly matched nothing; it
now matches everything. (fd.o #70799, Simon McVittie)
D-Bus 1.7.6 (2013-10-09)
==
The “CSI Shrewsbury” release.
Build-time configuration changes:
• Directory change notification via dnotify on Linux is no longer
supported; it hadn't compiled successfully since 2010 in any case.
If you don't have inotify (Linux) or kqueue (*BSD), you will need
to send SIGHUP to the dbus-daemon when its configuration changes.
(fd.o #33001, Chengwei Yang)
• Compiling with --disable-userdb-cache is no longer supported;
it didn't work since at least 2008, and would lead to an extremely
slow dbus-daemon even it worked. (fd.o #15589, #17133, #66947;
Chengwei Yang)
• The DBUS_DISABLE_ASSERTS CMake option didn't actually disable most
assertions. It has been renamed to DBUS_DISABLE_ASSERT to be consistent
with the Autotools build system. (fd.o #66142, Chengwei Yang)
• --with-valgrind=auto enables Valgrind instrumentation if and only if
valgrind headers are available. The default is still --with-valgrind=no.
(fd.o #56925, Simon McVittie)
Dependencies:
• Platforms with no 64-bit integer type are no longer supported.
(fd.o #65429, Simon McVittie)
• GNU make is now (documented to be) required. (fd.o #48277, Simon McVittie)
• Full test coverage no longer requires dbus-glib, although the tests do not
exercise the shared library (only a static copy) if dbus-glib is missing.
(fd.o #68852, Simon McVittie)
Enhancements:
• D-Bus Specification 0.22
· Document GetAdtAuditSessionData() and
GetConnectionSELinuxSecurityContext() (fd.o #54445, Simon)
· Fix example .service file (fd.o #66481, Chengwei Yang)
· Don't claim D-Bus is "low-latency" (lower than what?), just
give factual statements about it supporting async use
(fd.o #65141, Justin Lee)
· Document the contents of .service files, and the fact that
system services' filenames are constrained
(fd.o #66608; Simon McVittie, Chengwei Yang)
• Be thread-safe by default on all platforms, even if
dbus_threads_init_default() has not been called. For compatibility with
older libdbus, library users should continue to call
dbus_threads_init_default(): it is harmless to do so.
(fd.o #54972, Simon McVittie)
• Add GetConnectionCredentials() method (fd.o #54445, Simon)
• New API: dbus_setenv(), a simple wrapper around setenv().
Note that this is not thread-safe. (fd.o #39196, Simon)
• Add dbus-send --peer=ADDRESS (connect to a given peer-to-peer connection,
like --address=ADDRESS in previous versions) and dbus-send --bus=ADDRESS
(connect to a given bus, like dbus-monitor --address=ADDRESS).
dbus-send --address still exists for backwards compatibility,
but is no longer documented. (fd.o #48816, Andrey Mazo)
• Windows-specific:
· "dbus-daemon --nofork" is allowed on Windows again. (fd.o #68852,
Simon McVittie)
Fixes:
• Avoid an infinite busy-loop if a signal interrupts waitpid()
(fd.o #68945, Simon McVittie)
• Clean up memory for parent nodes when objects are unexported
(fd.o #60176, Thomas Fitzsimmons)
• Make dbus_connection_set_route_peer_messages(x, FALSE) behave as
documented. Previously, it assumed its second parameter was TRUE.
(fd.o #69165, Chengwei Yang)
• Escape addresses containing non-ASCII characters correctly
(fd.o #53499, Chengwei Yang)
• Document <servicedir> search order correctly (fd.o #66994, Chengwei Yang)
• Don't crash on "dbus-send --session / x.y.z" which regressed in 1.7.4.
(fd.o #65923, Chengwei Yang)
• If malloc() returns NULL in _dbus_string_init() or similar, don't free
an invalid pointer if the string is later freed (fd.o #65959, Chengwei Yang)
• If malloc() returns NULL in dbus_set_error(), don't va_end() a va_list
that was never va_start()ed (fd.o #66300, Chengwei Yang)
• fix build failure with --enable-stats (fd.o #66004, Chengwei Yang)
• fix a regression test on platforms with strict alignment (fd.o #67279,
Colin Walters)
• Avoid calling function parameters "interface" since certain Windows headers
have a namespace-polluting macro of that name (fd.o #66493, Ivan Romanov)
• Assorted Doxygen fixes (fd.o #65755, Chengwei Yang)
• Various thread-safety improvements to static variables (fd.o #68610,
Simon McVittie)
• Make "make -j check" work (fd.o #68852, Simon McVittie)
• Fix a NULL pointer dereference on an unlikely error path
(fd.o #69327, Sviatoslav Chagaev)
• Improve valgrind memory pool tracking (fd.o #69326,
Sviatoslav Chagaev)
• Don't over-allocate memory in dbus-monitor (fd.o #69329,
Sviatoslav Chagaev)
• dbus-monitor can monitor dbus-daemon < 1.5.6 again
(fd.o #66107, Chengwei Yang)
• Unix-specific:
· If accept4() fails with EINVAL, as it can on older Linux kernels
with newer glibc, try accept() instead of going into a busy-loop.
(fd.o #69026, Chengwei Yang)
· If socket() or socketpair() fails with EINVAL or EPROTOTYPE,
for instance on Hurd or older Linux with a new glibc, try without
SOCK_CLOEXEC. (fd.o #69073; Pino Toscano, Chengwei Yang)
· Fix a file descriptor leak on an error code path.
(fd.o #69182, Sviatoslav Chagaev)
· dbus-run-session: clear some unwanted environment variables
(fd.o #39196, Simon)
· dbus-run-session: compile on FreeBSD (fd.o #66197, Chengwei Yang)
· Don't fail the autolaunch test if there is no DISPLAY (fd.o #40352, Simon)
· Use dbus-launch from the builddir for testing, not the installed copy
(fd.o #37849, Chengwei Yang)
· Fix compilation if writev() is unavailable (fd.o #69409,
Vasiliy Balyasnyy)
· Remove broken support for LOCAL_CREDS credentials passing, and
document where each credential-passing scheme is used (fd.o #60340,
Simon McVittie)
· Make autogen.sh work on *BSD by not assuming GNU coreutils functionality
(fd.o #35881, #69787; Chengwei Yang)
· dbus-monitor: be portable to NetBSD (fd.o #69842, Chengwei Yang)
· dbus-launch: stop using non-portable asprintf (fd.o #37849, Simon)
· Improve error reporting from the setuid activation helper (fd.o #66728,
Chengwei Yang)
• Windows-specific:
· Remove unavailable command-line options from 'dbus-daemon --help'
(fd.o #42441, Ralf Habacker)
· Add support for looking up local TCPv4 clients' credentials on
Windows XP via the undocumented AllocateAndGetTcpExTableFromStack
function (fd.o #66060, Ralf Habacker)
· Fix insufficient dependency-tracking (fd.o #68505, Simon McVittie)
· Don't include wspiapi.h, fixing a compiler warning (fd.o #68852,
Simon McVittie)
• Internal changes:
· add DBUS_ENABLE_ASSERT, DBUS_ENABLE_CHECKS for less confusing
conditionals (fd.o #66142, Chengwei Yang)
· improve verbose-mode output (fd.o #63047, Colin Walters)
· consolidate Autotools and CMake build (fd.o #64875, Ralf Habacker)
· fix various unused variables, unusual build configurations
etc. (fd.o #65712, #65990, #66005, #66257, #69165, #69410, #70218;
Chengwei Yang, Vasiliy Balyasnyy)
D-Bus 1.7.4 (2013-06-13)
==
The “but is your thread-safety thread-safe?” release.
Security fixes:
• CVE-2013-2168: Fix misuse of va_list that could be used as a denial
of service for system services. Vulnerability reported by Alexandru Cornea.
(Simon)
Dependencies:
• The Windows version of libdbus now contains a C++ source file, used
to provide global initialization when the library is loaded.
gcc (mingw*) users should ensure that g++ is also installed.
• The libxml2-based configuration reader (which hasn't worked for 2.5 years,
and was never the recommended option) has been removed. Expat is now a
hard dependency.
Enhancements:
• It should now be safe to call dbus_threads_init_default() from any thread,
at any time. Authors of loadable modules and plugins that use libdbus
should consider doing so during initialization.
(fd.o #54972, Simon McVittie)
• Improve dbus-send documentation and command-line parsing (fd.o #65424,
Chengwei Yang)
Unix-specific:
· dbus-run-session: experimental new tool to start a temporary D-Bus
session, e.g. for regression tests or a text console, replacing
certain uses of dbus-launch which weren't really correct
(fd.o #39196, Simon)
Other fixes:
• In dbus-daemon, don't crash if a .service file starts with key=value
(fd.o #60853, Chengwei Yang)
• Unix-specific:
· Fix a crash similar to CVE-2013-2168 the first time we try to use syslog
on a platform not defining LOG_PERROR, such as Solaris or QNX.
This regressed in 1.7.0. (Simon)
· Fix an assertion failure if we try to activate systemd services before
systemd connects to the bus (fd.o #50199, Chengwei Yang)
· Avoid compiler warnings for ignoring the return from write()
(Chengwei Yang)
• Windows-specific:
· Under cmake, install runtime libraries (DLLs) into bin/ instead of lib/
so that Windows finds them (fd.o #59733, Ralf Habacker)
D-Bus 1.7.2 (2013-04-25)
==
The “only partially opaque” release.
Configuration changes:
• On non-QNX Unix platforms, the default limit on fds per message in the
session bus configuration has reduced from 4096 to 1024. The default
limit used on the system bus was already 1024. On QNX, both limits are
reduced further, to 128.
Enhancements:
• D-Bus Specification 0.21
· Following Unicode Corrigendum #9, the noncharacters U+nFFFE, U+nFFFF,
U+FDD0..U+FDEF are allowed in UTF-8 strings again. (fd.o #63072,
Simon McVittie)
Fixes:
• Diagnose incorrect use of dbus_connection_get_data() with negative slot
(i.e. before allocating the slot) rather than returning junk
(fd.o #63127, Dan Williams)
• Fix a cmake build regression since 1.7.0 (fd.o #63682; Ralf Habacker,
Simon McVittie)
• Unix-specific:
· On Linux, link successfully with glibc 2.17 (fd.o #63166, Simon McVittie)
· Under systemd, log to syslog only, not stderr, avoiding duplication
(fd.o #61399, #39987; Colin Walters, Dagobert Michelsen)
· Under systemd, remove unnecessary dependency on syslog.socket
(fd.o #63531, Cristian Rodríguez)
· Include alloca.h for alloca() if available, fixing compilation on
Solaris 10 (fd.o #63071, Dagobert Michelsen)
· Allow use of systemd-logind without the rest of systemd
(fd.o #62585, Martin Pitt)
· When built with CMake, link to librt and use the right path for
meinproc's XSLT stylesheets (fd.o #61637, Ralf Habacker)
· Reduce the default limit on number of fds per message to 128 under
QNX, working around an arbitrary OS limit (fd.o #61176, Matt Fischer)
• Windows-specific:
· Do not claim that all bus clients have the dbus-daemon's credentials;
pick up local TCPv4 clients' credentials (process ID and security
identifier, i.e. user) using GetExtendedTcpTable() (fd.o #61787,
Ralf Habacker)
D-Bus 1.7.0 (2013-02-22)
==
The "Disingenuous Assertions" release.
This is a new development release, starting the 1.7.x branch. D-Bus 1.6
remains the recommended version for long-term-supported distributions
or the upcoming GNOME 3.8 release.
Build-time configuration changes:
• The --with-dbus-session-bus-default-address configure option is no longer
supported. Use the new --with-dbus-session-bus-connect-address and
--with-dbus-session-bus-listen-address options instead. On Windows, you
usually want them to have the same argument; on Unix, the defaults are
usually correct.
• Similarly, the DBUS_SESSION_BUS_DEFAULT_ADDRESS CMake variable is no longer
supported; use the new DBUS_SESSION_BUS_LISTEN_ADDRESS and
DBUS_SESSION_BUS_CONNECT_ADDRESS variables instead.
• cmake/cross-compile.sh has been removed. Instead, please use a
cross-toolchain file (-DCMAKE_TOOLCHAIN_FILE) as documented at
<http://www.vtk.org/Wiki/CMake_Cross_Compiling>; or use Autotools
as documented in "info automake Cross-Compilation", and set
PKG_CONFIG_PATH appropriately.
Requirements:
• Man pages now require xmlto (or either xmlto or meinproc, if using CMake).
• man2html is no longer used.
Enhancements:
• D-Bus Specification 0.20
· actually say that /org/freedesktop/DBus is the object that
implements o.fd.DBus (fd.o #51865, Colin Walters)
· various reorganisation for better clarity (fd.o #38252, Simon McVittie)
· stop claiming that all basic types work just like INT32 (strings don't!)
• The "source code" for the man pages is now Docbook XML, eliminating
the outdated duplicate copies used when building with CMake.
(fd.o #59805; Ralf Habacker, Simon McVittie)
Fixes:
• In the activation helper, when compiled for tests, do not reset the system
bus address, fixing the regression tests. (fd.o #52202, Simon)
• Fix building with Valgrind 3.8, at the cost of causing harmless warnings
with Valgrind 3.6 on some compilers (fd.o #55932, Arun Raghavan)
• Merge <servicehelper> from system-local.conf if necessary (fd.o #51560,
Krzysztof Konopko)
• Under CMake, prefer xmlto over meinproc (fd.o #59733, Ralf Habacker)
• Stop duplicating CMake's own logic to find libexpat
(fd.o #59733, Ralf Habacker)
• Don't assume CMake host and build system are the same (fd.o #59733,
Ralf Habacker)
• Avoid deprecation warnings for GLib 2.35 (fd.o #59971, Simon McVittie)
• Unix-specific:
· Check for functions in libpthread correctly, fixing compilation on
(at least) OpenBSD (fd.o #47239, Simon)
· Don't leak temporary fds pointing to /dev/null (fd.o #56927,
Michel HERMIER)
· Update sd-daemon.[ch] from systemd (fd.o #60681)
· Add partial support for QNX (fd.o #60339, fd.o #61176; Matt Fischer)
• Windows-specific:
· The default session bus listening and connecting address is now
"autolaunch:", which makes D-Bus on Windows interoperate with itself
and GDBus "out of the box". Use the configure options and cmake variables
described above if you require a different autolaunch scope.
(fd.o #38201, Simon McVittie)
· Avoid a CMake warning under Cygwin (fd.o #59401, Ralf Habacker)
• Create session.d, system.d directories under CMake (fd.o #41319,
Ralf Habacker)
D-Bus 1.6.18 (2013-11-01)
==
The “sunflower plains” release.
Fixes:
• path_namespace='/' in match rules incorrectly matched nothing; it
now matches everything. (fd.o #70799, Simon McVittie)
D-Bus 1.6.16 (2013-10-08)
==
The “Fortify Agility” release.
Fixes:
• Make dbus_connection_set_route_peer_messages(x, FALSE) behave as
documented. Previously, it assumed its second parameter was TRUE.
(fd.o #69165, Chengwei Yang)
• Fix a NULL pointer dereference on an unlikely error path
(fd.o #69327, Sviatoslav Chagaev)
• Unix-specific:
· If accept4() fails with EINVAL, as it can on older Linux kernels
with newer glibc, try accept() instead of going into a busy-loop.
(fd.o #69026, Chengwei Yang)
· If socket() or socketpair() fails with EINVAL or EPROTOTYPE,
for instance on Hurd or older Linux with a new glibc, try without
SOCK_CLOEXEC. (fd.o #69073; Pino Toscano, Chengwei Yang)
· Fix a file descriptor leak on an error code path.
(fd.o #69182, Sviatoslav Chagaev)
· Fix compilation if writev() is unavailable (fd.o #69409,
Vasiliy Balyasnyy)
D-Bus 1.6.14 (2013-09-05)
==
The “Restore Fatigue” release.
Fixes:
• Avoid an infinite busy-loop if a signal interrupts waitpid()
(fd.o #68945, Simon McVittie)
• Escape addresses containing non-ASCII characters correctly
(fd.o #53499, Chengwei Yang)
• If malloc() returns NULL in _dbus_string_init() or similar, don't free
an invalid pointer if the string is later freed (fd.o #65959, Chengwei Yang)
• If malloc() returns NULL in dbus_set_error(), don't va_end() a va_list
that was never va_start()ed (fd.o #66300, Chengwei Yang)
• Fix a regression test on platforms with strict alignment (fd.o #67279,
Colin Walters)
• Avoid calling function parameters "interface" since certain Windows headers
have a namespace-polluting macro of that name (fd.o #66493, Ivan Romanov)
• Make "make -j check" work (fd.o #68852, Simon McVittie)
D-Bus 1.6.12 (2013-06-13)
==
Fixes:
• CVE-2013-2168: Fix misuse of va_list that could be used as a denial
of service for system services. Vulnerability reported by Alexandru Cornea.
(Simon)
• In dbus-daemon, don't crash if a .service file starts with key=value
(fd.o #60853, Chengwei Yang)
• Unix-specific:
· Fix an assertion failure if we try to activate systemd services before
systemd connects to the bus (fd.o #50199, Chengwei Yang)
· Avoid compiler warnings for ignoring the return from write()
(Chengwei Yang)
D-Bus 1.6.10 (2013-04-24)
==
The “little-known facts about bananas” release.
• Following Unicode Corrigendum #9, the noncharacters U+nFFFE, U+nFFFF,
U+FDD0..U+FDEF are allowed in UTF-8 strings again.
(fd.o #63072, Simon McVittie)
• Diagnose incorrect use of dbus_connection_get_data() with negative slot
(i.e. before allocating the slot) rather than returning junk
(fd.o #63127, Dan Williams)
• In the activation helper, when compiled for tests, do not reset the system
bus address, fixing the regression tests. (fd.o #52202, Simon)
• Fix building with Valgrind 3.8, at the cost of causing harmless warnings
with Valgrind 3.6 on some compilers (fd.o #55932, Arun Raghavan)
• Don't leak temporary fds pointing to /dev/null (fd.o #56927, Michel HERMIER)
• Create session.d, system.d directories under CMake (fd.o #41319,
Ralf Habacker)
• Unix-specific:
· Include alloca.h for alloca() if available, fixing compilation on
Solaris 10 (fd.o #63071, Dagobert Michelsen)
D-Bus 1.6.8 (2012-09-28)
==
The "Fix one thing, break another" release.
• Follow up to CVE-2012-3524: The additional hardening
work to use __secure_getenv() as a followup to bug #52202
broke certain configurations of gnome-keyring. Given
the difficulty of making this work without extensive
changes to gnome-keyring, use of __secure_getenv() is
deferred.
D-Bus 1.6.6 (2012-09-28)
==
The "Clear the environment in your setuid binaries, please" release.
• CVE-2012-3524: Don't access environment variables (fd.o #52202)
Thanks to work and input from Colin Walters, Simon McVittie,
Geoffrey Thomas, and others.
• Unix-specific:
· Fix compilation on Solaris (fd.o #53286, Jonathan Perkin)
· Work around interdependent headers on OpenBSD by including sys/types.h
before each use of sys/socket.h (fd.o #54418, Brad Smith)
this is a step of two major branches, to much to list here
pkgsrc notes:
-I've left out our NETBSD_ATOMIC_OPS patch because it is very invasive
and gcc intrinsics are hopefully good enough
-A workaround for non-standard behavior of vsnprintf has been added
upstream, for HPUX as the comment says. We had a patch for __hpux||__sgi
which is removed now - the sgi case should be checked.
This changes the buildlink3.mk files to use an include guard for the
recursive include. The use of BUILDLINK_DEPTH, BUILDLINK_DEPENDS,
BUILDLINK_PACKAGES and BUILDLINK_ORDER is handled by a single new
variable BUILDLINK_TREE. Each buildlink3.mk file adds a pair of
enter/exit marker, which can be used to reconstruct the tree and
to determine first level includes. Avoiding := for large variables
(BUILDLINK_ORDER) speeds up parse time as += has linear complexity.
The include guard reduces system time by avoiding reading files over and
over again. For complex packages this reduces both %user and %sys time to
half of the former time.
