Use epoll in a backwards-compatible way on Linux < 2.6.27

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33337
Bug-NB: NB#197191
Bug-NB: NB#225019

Maintain the set of polled sockets over time

This enables us to benefit from epoll(4): polling is now
O(number of watch changes + number of active fds), whereas with poll(2)
it's O(total number of fds).

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33337
Bug-NB: NB#197191

Add a stub _dbus_loop_toggle_watch and call it where needed

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33337
Bug-NB: NB#197191

Add an implementation of DBusSocketSet using epoll

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33337
Bug-NB: NB#197191
Bug-NB: NB#225019

Check for epoll in configure.ac

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33337
Bug-NB: NB#197191

DBusSocketSet: new abstraction for struct pollfd[] or whatever

In this second version of this patch, DBusSocketSet is an "abstract base
class" so that when using a better OS-specific API fails, we can always fall
back to _dbus_poll(). For instance, this is necessary when the "better
OS-specific API" is epoll on Linux, the build machine has a modern glibc
version, and the host machine either has an old kernel, is emulated in qemu
(which does not support the epoll syscalls yet), or both.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33337
Bug-NB: NB#197191
Bug-NB: NB#225019

NEWS

DBusLoop: store watches in a hash table keyed by fd

This means we don't need to build up the watches_for_fds array, because
it's quick to find the watches based on the fd.

This also fixes fd.o #23194, because we only put each fd in the
array of DBusPollFD once, with the union of all its watches' flags; we
can no longer get more entries in the array than our number of file
descriptors, which is capped at our RLIMIT_NOFILE, so we can't get
EINVAL from poll(2).

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=23194
Reviewed-by: Thiago Macieira <thiago@kde.org>

DBusLoop: fold Callback into TimeoutCallback

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33342
Reviewed-by: Thiago Macieira <thiago@kde.org>

Drop WatchCallback entirely, just use a list of DBusWatch in the main loop

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33342
Reviewed-by: Thiago Macieira <thiago@kde.org>

DBusLoop: inline add_callback, remove_callback into their callers

The watch and timeout code paths will diverge completely when we change
WatchCallback * to just be a DBusWatch *, removing the benefit of having
common code.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33342
Reviewed-by: Thiago Macieira <thiago@kde.org>

DBusLoop: move OOM flag in watches inside the DBusWatch

This will eventually let us maintain a DBusPollFD[] of just the active
watches, between several iterations. The more immediate benefit is that
WatchCallback can go away, because it only contains a refcount, a
now-useless type, and a watch that already has its own refcount.

This doesn't take any more memory for DBusWatch when not using DBusLoop
(e.g. in client/service code or bindings), because we're just using more
bits in an existing bitfield.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33342
Reviewed-by: Thiago Macieira <thiago@kde.org>

DBusLoop: keep separate lists of watches and timeouts

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33342
Reviewed-by: Thiago Macieira <thiago@kde.org>

DBusLoop: factor out watch_flags_to_poll_events, watch_flags_from_poll_revents

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33342
Reviewed-by: Thiago Macieira <thiago@kde.org>

Remove _dbus_loop_add_watch_full

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33342
Reviewed-by: Thiago Macieira <thiago@kde.org>

bus-activation: separate the "finished" callback from the watch callback

This has been marked as broken since 2003...

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33342
Reviewed-by: Thiago Macieira <thiago@kde.org>

DBusLoop: remove second layer of watch callbacks where possible

Similar to the previous commit, almost every use of DBusWatch can just
have the main loop call dbus_watch_handle.

The one exception is the bus activation code; it's had a comment
explaining why it's wrong since 2003. We should fix that one day, but for
now, just migrate it to a new _dbus_loop_add_watch_full which preserves
the second-layer callback.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33342
Reviewed-by: Thiago Macieira <thiago@kde.org>

DBusLoop: remove a layer of pointless abstraction around timeouts

Instead of supplying 8 tiny wrapper functions around dbus_timeout_handle,
each with a user_data parameter that's a potentially unsafe borrowed
pointer but isn't actually used, we can call dbus_timeout_handle directly
and save a lot of trouble.

One of the wrappers previously called dbus_timeout_handle repeatedly
if it returned FALSE to indicate OOM, but that timeout's handler never
actually returned FALSE, so there was no practical effect. The rest just
ignore the return, which is documented as OK to do.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33342
Reviewed-by: Thiago Macieira <thiago@kde.org>

DBusConnection: ref the connection in the timeout handler

client_timeout_callback in bus/test.c refs the connection across the
timeout invocation, which looks suspiciously like a workaround. If we
make the timeout handler itself ref the connection, we won't need that,
and can simplify timeout handling drastically.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33342
Reviewed-by: Thiago Macieira <thiago@kde.org>

bus signal_handler: don't use _dbus_warn, and don't pretend to be portable

_dbus_warn isn't async-signal-safe, so that's out. We can use write()
instead; it's POSIX but not ISO C, but then again, so are signals.
Accordingly, guard it with DBUS_UNIX.

dbus-sysdeps-util-win doesn't actually implement _dbus_set_signal_handler
anyway, so not compiling this code on non-Unix seems more honest.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33336
Reviewed-by: Thiago Macieira <thiago@kde.org>

bus: signal_handler: ignore failure to write, and explain why

See the comment in the source code for rationale.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33336
Reviewed-by: Thiago Macieira <thiago@kde.org>

Check that watches are removed, then invalidated, then unreffed

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33336
Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Reviewed-by: Thiago Macieira <thiago@kde.org>

Always remove, invalidate and free watches before closing watched sockets

This should mean we don't get invalid fds in the main loop.

The BSD (kqueue) and Windows code paths are untested, but follow the same
patterns as the tested Linux/generic Unix versions.

DBusTransportSocket was already OK (it called free_watches() before
_dbus_close_socket, and that did the remove, invalidate, unref dance).

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33336
Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Reviewed-by: Thiago Macieira <thiago@kde.org>

bump version

ahem.

Prepare 1.5.4

Merge branch 'dbus-1.4'

Conflicts:
NEWS
configure.ac

version bump

prepare 1.4.12

fix installable tests to not install if unwanted

Merge branch 'dbus-1.4'

NEWS

Fix copyright dates

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=34570

Use DBUS_TIMEOUT_INFINITE in dbus-daemon.c

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=34570

Add and use DBUS_TIMEOUT_INFINITE and DBUS_TIMEOUT_USE_DEFAULT

The documentation claimed that INT_MAX (whatever that means) meant the
default, but the value that has actually always been checked for is
0x7fffffff (aka INT32_MAX on the competent platforms we sadly don't
restrict our portability to), so we should use that. (In practice D-Bus
probably never worked on platforms where int wasn't 32 bits, though.)

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=34570

Test that a message with the byte order mangled causes disconnection but no crash

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=38120

marshal test: test dbus_message_demarshal_bytes_needed

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=38120

Add a test for marshalling and endian-swapping

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=38120

installcheck: don't run installed tests against installed library if in a DESTDIR

That probably won't work, because it'll find the system-wide library
which might be older.

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=34570

Alter test-dbus-daemon so it tests the installed dbus-daemon by default

For installcheck, adjust it to use things from DESTDIR.

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=34570

Add support for installing most of the modular tests

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=34570

new test for being disconnected by a corrupt message stream

This was meant to reproduce fd.o #15578, but doesn't; the cause of that
bug isn't this simple.

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=34570

modular tests: depend on GIO 2.22 for GSocket

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=34570

Test nonce-tcp transport

Regression test for https://bugs.freedesktop.org/show_bug.cgi?id=34569

Reviewed-by: David Zeuthen <davidz@redhat.com>
Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=34570

dbus-daemon test: sanity-check the just-installed config file when using DESTDIR

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=34570

Add a simple integration test for dbus-daemon

This just pushes 2000 messages (or 100000 in performance-testing mode)
through the dbus-daemon, to an echo service and back.

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=34570

Run integration tests on the installed dbus binaries during installcheck

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=34570

Give the tests DBUS_TEST_DAEMON and DBUS_TEST_DATA in their environment

This will allow modular tests to spawn a dbus-daemon with a specified
config file; nothing uses this just yet.

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=34570

Attempt to reproduce fd.o #34393 via another regression test

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=34570

loopback test: unref messages after use

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=34570

Add an end-to-end sanity check for TCP and Unix DBusServer/DBusConnection

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=34570

Add support for building "modular" tests, which require GLib and dbus-glib

For the moment, the CMake build system only knows about the existing
"embedded tests"; make it define both symbols, though.

We use GLib because it has GTester (and life's too short to write yet another
JUnit clone), and dbus-glib for the main-loop integration only (see
fd.o #31515 for thoughts on incorporating just those two functions in a
separate library in the dbus tarball).

I'm not using DBusLoop for the main loop because I specifically don't
want to use non-public API or ABI of libdbus in the modular tests. If we make
sure they work against a shared libdbus, we can use them to test the
installed system, with "make installcheck".

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=34570

Don't disable GLib assertions when disabling our own assertions

We no longer use GLib internally, and assertions are how it'll report test
failures when we add GTest-based tests.

Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=34570

dbus_message_demarshal_bytes_needed: correct a wrong assertion

It's entirely possible for a message to indicate how many bytes we need,
without actually being complete.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=38120
Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>

_dbus_header_byteswap: change the first byte of the message, not just the struct member

This has been wrong approximately forever, for instance see:
http://lists.freedesktop.org/archives/dbus/2007-March/007357.html

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=38120
Bug-Debian: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=629938
Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>

Consistently use atomic operations for all access to DBusConnection refcount

Trying to mix atomic operations with locked non-atomic operations is
broken: the atomic ops aren't necessarily atomic with respect to the
locked non-atomic ops, and the non-atomic ops aren't protected by the
lock because the atomic ops can change the refcount behind their back.

In theory we could use the connection lock if atomic ops aren't supported
(making a per-connection lock cheaper than the global lock used to
implement atomic ops) *and* our mutexes are recursive (making it safe
against deadlocks)... but life's too short.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=38005
Tested-by: Will Manley <freedesktop williammanley net>
Reviewed-by: Will Thompson <will.thompson@collabora.co.uk>

Merge branch 'dbus-1.4'

NEWS

bus: use ln -fs to enable dbus in systemd, not $(LN_S)

Using $(LN_S) is inappropriate because it could in theory mean either
ln -s, ln or cp -p depending on autoconf checks.

Not using -f breaks reinstallation directly from source (DESTDIR unset),
because the symlinks will already exist.

Because systemd isn't currently portable to non-Linux, let alone
non-SUS-compliant systems, it seems safe to assume that ln -fs behaves
as specified by SUS if systemd was found.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=37870
Reviewed-by: Colin Walters <walters@verbum.org>

Merge branch 'dbus-1.4'

Conflicts:
NEWS

NEWS

Remove support for -Wl,--gc-sections altogether

Packagers should only enable this flag if they have confirmed that it
actually works on their toolchain (it's the sort of rarely used feature
that frequently regresses on obscure architectures/OSs without anyone
noticing), and also confirmed that it is actually a significant size win
for their configuration.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33466
Reviewed-by: Colin Walters <walters@verbum.org>

don't try to build dbus-daemon-launch-helper for Windows

Fixed regression in b0acded which breaks build for Windows.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=37838
Reviewed-by: Simon McVittie <simon.mcvittie@collabora.co.uk>

Use #!/bin/sh for run-with-tmp-session-bus.sh

It doesn't seem to have any bashisms, and bash isn't in /bin on *BSD.

Suggested by Timothy Redaelli.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=35880

add a release name here too

add release name :-)

restart development

Finalize D-Bus Spec 0.17, update lists of authors

Merge branch 'dbus-1.4'

Conflicts:
NEWS
configure.ac

Development version

Prepare version 1.4.10

Fixes for Visual Studio 2010 cmake generator - reported by Ray Satiro.

Fixed cmake syntax errors.

Merge branch 'dbus-1.4'

NEWS

bus_context_log: divert messages to stderr if we're not using syslog

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=35358
Reviewed-by: Colin Walters <walters@verbum.org>

process_config_first_time_only: initialize syslog as intended.

On Linux, we previously called openlog() (with different options!) while
initializing SELinux; leave SELinux messages as LOG_USER|LOG_INFO in case
anyone was relying on that, but let the rest of our log messages come
out as LOG_DAEMON.

Also enable LOG_PERROR (copy syslog messages to stderr) unconditionally;
we can make this an autoconf check if anyone's pet Unix doesn't have
LOG_PERROR.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=35358
Reviewed-by: Colin Walters <walters@verbum.org>

_dbus_init_system_log: remove FIXME on Windows

We don't need any initialization here, so there's nothing to fix.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=35358
Reviewed-by: Colin Walters <walters@verbum.org>

bus_context_check_security_policy: syslog if we hit the outgoing quota

In practice, nothing copes with missing broadcast signals, so the least
we can do is make the failure mode visible.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=35358
Reviewed-by: Colin Walters <walters@verbum.org>

bus_context_check_security_policy: for SELinux denials, share code to set errors

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=35358
Reviewed-by: Colin Walters <walters@verbum.org>

bus_context_check_security_policy: factor out complain_about_message

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=35358
Reviewed-by: Colin Walters <walters@verbum.org>

Merge branch 'dbus-1.4'

Check for X even if X11 autolaunching is disabled

DBUS_ENABLE_X11_AUTOLAUNCH obviously requires DBUS_BUILD_X11. However,
the converse is not true.

If DBUS_BUILD_X11 is defined, dbus-launch will be able to connect to
the X server to determine when the session ends; most distributors will
want this, but it can be disabled with the standard Autoconf option
--without-x.

If DBUS_ENABLE_X11_AUTOLAUNCH is *also* defined, dbus-launch and libdbus
will be willing to perform autolaunch. Again, most distributors will want
this, but it can be disabled with --disable-x11-autolaunch.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=19997
Reviewed-by: Colin Walters <walters@verbum.org>

test-autolaunch: don't expect autolaunching to work if X11 is disabled

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=19997
Reviewed-by: Colin Walters <walters@verbum.org>

Merge branch 'dbus-1.4'

NEWS

When checking for __sync_sub_and_fetch, don't underquote, to shut up recent autoconf

Without the correct number of levels of quoting, autoconf mistakenly
believes we didn't use AC_LANG_SOURCE where required. (In fact,
AC_LANG_PROGRAM calls AC_LANG_SOURCE.)

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=19681
Reviewed-by: Colin Walters <walters@verbum.org>

When checking for va_copy, use AC_LANG_SOURCE to shut up recent autoconf

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=19681
Reviewed-by: Colin Walters <walters@verbum.org>

Use TP_ADD_COMPILER_FLAG to simplify application of warning-like CFLAGS

This also means we check for support for them.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=19681
Reviewed-by: Colin Walters <walters@verbum.org>

Use TP_COMPILER_WARNINGS for all -Wfoo options

This consistently checks whether all these options actually work in the
current version of gcc.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=19681
Reviewed-by: Colin Walters <walters@verbum.org>

Import tp-compiler-flag.m4 and tp-compiler-warnings.m4 from telepathy-glib

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=19681
Reviewed-by: Colin Walters <walters@verbum.org>

NEWS

Merge branch 'dbus-1.4'

NEWS

handle_server_data_anonymous_mech: remove unnecessary debug output

Doing a malloc and a hex-encoding pass just to produce a _dbus_verbose
message (i.e. a message that, in practice, nobody will see) seems like
overkill, and this block had incorrect error handling (not checking the
result of _dbus_string_init) which upsets static analysis tools.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=29881
Bug-NB: NB#180486
Reviewed-by: Cosimo Alfarano <cosimo.alfarano@collabora.co.uk>

dbus-launch: pass_info: always free strings on OOM

This doesn't really do anything, because we're about to exit anyway, but
it placates static analysis tools.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=29881
Bug-NB: NB#180486
Reviewed-by: Cosimo Alfarano <cosimo.alfarano@collabora.co.uk>

dbus-send: remove minor dead code

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33128
Reviewed-by: Cosimo Alfarano <cosimo.alfarano@collabora.co.uk>

dbus_connection_dispatch: remove dead code

There's no way pending can be non-NULL here; if it was, we'd have jumped
straight past this block (getting filters from the connection), because
replies to pending calls don't go through filters.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=33128
Reviewed-by: Cosimo Alfarano <cosimo.alfarano@collabora.co.uk>

Spelling fix.

Definition is obsolate with commit e0fc90bdf2c60a4c77fffaf490d6f06b2bd6999d