ci: Reshuffle mingw jobs so we test different combinations

We test the combinations that we don't test on Travis-CI.

Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit a6d926b805a1a38567a70490b3e8d7d6c932d1f5)

ci: Use ccache to speed up repeated builds

Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit d0728fd06e5a2302e7596e3df56b68b0a0834fd7)

ci: Add Gitlab-CI configuration

This uses the same shell scripts as Travis-CI, with slightly different
settings. We use Docker containers for all our Gitlab-CI runs, so take
the opportunity to use Debian 9 'stretch' as our baseline, and
relegate Ubuntu 14.04 'trusty' to to a secondary build.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=108177
Acked-by: Philip Withnall <withnall@endlessm.com>
(cherry picked from commit 60933c09e9e891f74f0102fabe22d29a1a7ae5c5)

ci: Explicitly install cmake

Travis-CI workers have cmake preinstalled, but Gitlab-CI Docker images
typically don't.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=108177
Acked-by: Philip Withnall <withnall@endlessm.com>
(cherry picked from commit 907832e00849ca454322052981dbb122ea537506)

ci: Teach ci-install.sh to install wine on Debian 9 'stretch'

Signed-off-by: Simon McVittie <smcv@collabora.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=108177
Acked-by: Philip Withnall <withnall@endlessm.com>
(cherry picked from commit 408b222a9fc61327cd7be385b6705f30f0c38802)

travis-ci: Add cross building support for mingw 64 bit compiler

Signed-off-by: Ralf Habacker <ralf.habacker@freenet.de>
Reviewed-by: Simon McVittie <smcv@collabora.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=105662
(cherry picked from commit d22e7901b555a0bfb5e06fb2463d839a276c7482)

sysdeps-win: Print word-size-dependent offset correctly

AddrPC.Offset is the same size as a pointer, but previously
we printed it as though it was the same size as a long,
which is 32 bits on 64-bit Windows.

Reviewed-by: Simon McVittie <smcv@collabora.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=105662
(cherry picked from commit 577813cf3a89df804efa6d85a1c5415ba12806ec)

dbus-transport-socket: Correctly print DBusSocket with DBUS_SOCKET_FORMAT

Previously, on 64-bit Windows we were passing a 32-bit int where the
format string expects a 64-bit SOCKET.

Reviewed-by: Simon McVittie <smcv@collabora.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=105662
(cherry picked from commit 18d4ff664491c17664b9f88c06d9338cd3750120)

Update NEWS

Reference the freedesktop.org Code of Conduct

Signed-off-by: Simon McVittie <smcv@collabora.com>

Update NEWS

Do not apply __attribute__((__malloc__)) to dbus_realloc()

As noted in GLib commit c879f50f, gcc's interpretation of the malloc
attribute has become more strict over time, which could result in
miscompilation. The new definition is that in addition to assuming
that the returned memory block is newly-allocated, gcc now assumes
that it does not contain any valid pointers. This is OK for
uninitialized or zero-initialized memory returned by dbus_malloc()
or dbus_malloc0(), but not valid for dbus_realloc(), which might be
used for a dynamically-sized array of (structures containing)
valid pointers.

See https://gitlab.gnome.org/GNOME/glib/issues/1465

Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=107741

Update NEWS

server-unix: Don't leak address of systemd server on success

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=107320
Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
(cherry picked from commit d98c43c697fbeb104463dcf2da36d0d855bfb367)

bus: Free address (from --address) when we have finished using it

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=107320
Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
(cherry picked from commit 7ae750c4e887b2d63f87caaf2162125ec3217fcc)

1.12.11

1.12.10

Update NEWS

validate_body_helper: Bounds-check before validating booleans

Running the "embedded tests" through valgrind revealed that before this
commit, we would have been willing to read up to 3 bytes off the end of
a message if the message is truncated part way through a boolean. Any
practical allocator will round up allocations to the next 32-bit (or
larger) boundary, so in practice this will not leave the memory buffer
(and in particular did not crash during unit testing), but it could read
uninitialized contents.

On little-endian CPUs, an attacker might be able to use this to learn
whether up to 3 bytes of uninitialized memory in the dbus-daemon
were all-zero (their crafted message would be relayed) or not (their
connection would be disconnected for sending an invalid message). On
big-endian CPUs, an attacker might be able to use this to learn whether
up to 3 bytes were all-zeroes (relayed to a cooperating peer), 0-2
bytes of all-zeroes followed by 0x01 (relayed to a cooperating peer),
or something else (disconnected). This is not believed to be exploitable
to leak interesting information.

Fixes: 62e46533 "hardcode dbus_bool_t to 32 bits"
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=107332
Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Thiago Macieira <thiago@kde.org>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
(cherry picked from commit e93a775e68daeda5c95984452aee6327e31c17dd)

Update NEWS

nonce: Don't try to rmdir(NULL) on OOM

If re-initializing the string fails, it will be left in a state
where it has a length of 0 and a NULL buffer. That's valid to
"free", but not valid to pass to rmdir().

Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=107194
(cherry picked from commit 294e8b0b672c8ffdcb8d9227f114846433659864)

dbus_server_listen: Don't leak first_connect_error

If an implementation fails to listen, and a subsequent implementation
succeeds, then we would have leaked this. Detected by running
tests/loopback.c under valgrind.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=107194
(cherry picked from commit b14a4517a82f7e7e4c0b59cb663ebf77563decda)

sysdeps: Reassure gcc 8 that we are not overflowing struct sockaddr_un

Using strncpy (buffer, str, strlen (str)) is a "code smell" that
might indicate a serious bug (it effectively turns strncpy into
strcpy), and gcc 8 now warns about it. In fact we avoided the bug
here, but it wasn't at all obvious.

We already checked that path_len is less than or equal to
_DBUS_MAX_SUN_PATH_LENGTH, which is 99, chosen to be strictly less
than the POSIX minimum sizeof(sun_path) >= 100, so we couldn't
actually be overflowing the available buffer.

The new static assertion in this commit matches a comment above the
definition of _DBUS_MAX_SUN_PATH_LENGTH: we define
_DBUS_MAX_SUN_PATH_LENGTH to 99, because POSIX says struct
sockaddr_un's sun_path member is at least 100 bytes (including space
for a \0 terminator). dbus will now fail to compile on
platforms that are non-POSIX-compliant in this way, except for Windows.

We zeroed the struct sockaddr_un before writing into it, so stopping
one byte short of the end of sun_path ensures that we get \0
termination.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=107350
Reviewed-by: Thiago Macieira <thiago@kde.org>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
(cherry picked from commit f429631365ba59a1749438af2184cab138a31772)

build: Disable new gcc 8 warning -Wcast-function-type

The foreach(list, (DBusForeachFunction) free, NULL) idiom seems too
entrenched to remove it from stable branches.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=107349
Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Thiago Macieira <thiago@kde.org>

Update NEWS

Signed-off-by: Simon McVittie <smcv@collabora.com>

test: Skip TCP tests if getaddrinfo doesn't work

For example, this can be the case in bubblewrap or Debian pbuilder after
unsharing the network namespace:

    bwrap \
    --bind / / \
    --dev-bind /dev /dev \
    --bind /dev/shm /dev/shm \
    --bind /dev/pts /dev/pts \
    --unshare-net \
    ${builddir}/test/test-loopback --tap
    ...
    ok 1 /connect/tcp # SKIP Name resolution does not work here:
    getaddrinfo("127.0.0.1", "0", {flags=ADDRCONFIG, family=INET,
    socktype=STREAM, protocol=TCP}): Name or service not known

On some systems this can be circumvented by using nss_wrapper from
<https://cwrap.org/nss_wrapper.html>:

    cat > hosts <<EOF
    127.0.0.1 localhost
    EOF
    bwrap \
    ... \
    env \
    LD_PRELOAD=libnss_wrapper.so \
    NSS_WRAPPER_HOSTS=$(pwd)/hosts \
    ${builddir}/test/test-loopback --tap
    ...
    # listening at tcp:host=127.0.0.1,port=39219,family=ipv4,guid=...

but for systems where that does't work, we should be prepared to skip
the affected tests.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=106812
(cherry picked from commit f1faafd59bec67d347edb10447c4b9b18193458c)

server-oom test: Don't assume localhost is resolvable

Pathological autobuilder environments might not list localhost in
/etc/hosts.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=106812
(cherry picked from commit 4cfc7de30de7111f589501e5b597063aeb96cf28)

test: Test the same things with unix: that we do with tcp:

Minimal autobuilder environments don't always have working TCP,
so we may need to skip TCP tests. Make sure we test the equivalent
code paths via Unix sockets in those environments.

One notable exception is test/fdpass.c, which uses TCP as a transport
that is known not to be able to carry Unix fds; this needs to continue
to use TCP.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=106812
(cherry picked from commit cb7dd5bfccb2882d0e9951c41040ff6a97bb827d)

server-oom test: Parse the address instead of going directly to TCP

This expands test coverage, and lets us reuse the test for other
address schemes.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=106812
(cherry picked from commit b19c9e2f265878801bc48866a7cc3152ca2ad45e)

sysdeps-unix: Handle errors from getaddrinfo correctly

getaddrinfo and getnameinfo have their own error-handling convention
in which the library call returns either 0 or an EAI_* error code
unrelated to errno. If the error code is not EAI_SYSTEM, then
the value of errno is undefined (in particular it might be carried
over from a previous system call or library call). Introduce a
new helper function _dbus_error_from_gai() to handle this.

The equivalent code paths in Windows appear to be OK: the Windows
implementation of getaddrinfo() is documented to return a Winsock
error code, which we seem to be handling correctly.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=106395
(cherry picked from commit 60cedd0cfd775c9fcf7260e12af9b2ffeefc2bbe)

Start 1.12.10 development

Signed-off-by: Simon McVittie <smcv@collabora.com>

1.12.8

build: Uninstall JavaScript and CSS from htmldir

Otherwise, distcheck fails when mallard-ducktype is available.

Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit 9391d769ae335872e5b770f6741855fde45b8186)

Preallocate release name

NEWS: Mention non-local TCP too

Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit 2390a325a0f094a87065e6dcfdf99c188a5b360f)

Update NEWS

(cherry picked from commit ee0e42ae2d157a5d4c9d5f1a9114632897bc47ad)

dbus-daemon(1): Mention and deprecate shared session buses

This might (?) have made sense behind a firewall in 2003; but now it's
2018, the typical threat model that we are defending against has
changed from "vandals want to feel proud of their l33t skills"
to "organised crime wants your money", and a "trusted" local LAN
probably contains an obsolete phone, tablet, games console or
Internet-of-Things-enabled toaster with remote root exploits.
This make network topologies that used to be acceptable look
increasingly irresponsible.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=106004
Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
(cherry picked from commit d0a16b59a8572fbd1934e941e2e3004840306222)

dbus-daemon(1): Recommend requiring EXTERNAL on non-Windows OSs

This is the default, and blocks TCP-based attacks by making the
attacker fail to authenticate (while also preventing inadvisable
TCP-based configurations from working).

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=106004
Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Ralf Habacker <ralf.habacker@freenet.de>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
(cherry picked from commit aef4475939a773e1a205a71d641ea2bb6793ab92)

dbus-daemon(1): Put some scary warnings on <allow_anonymous/>

I'm far from convinced that this option should even *exist*, but it
should definitely be documented as a very bad thing.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=106004
Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Ralf Habacker <ralf.habacker@freenet.de>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
(cherry picked from commit 5d3680486712891c13b85c07fab629bb70f623cc)

dbus-daemon(1): Recommend against remote TCP for debugging

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=106004
Reviewed-by: Ralf Habacker <ralf.habacker@freenet.de>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
[smcv: Add a TODO comment as suggested]
Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit cf47380641aac0e5b40694b0ae09ffb85ec4b5fc)

dbus-daemon(1): Say that non-local TCP is insecure

With some fairly reasonable threat models (active or passive local
attacker able to eavesdrop on the network link, confidential
information being transferred via D-Bus), secure authentication is
insufficient to make this transport secure: it does not protect
confidentiality or integrity either.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=106004
Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Ralf Habacker <ralf.habacker@freenet.de>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
(cherry picked from commit 2513f84db68a9edad8558806b777ed6c284016b9)

Update NEWS for 1.12.x branch

Signed-off-by: Simon McVittie <smcv@collabora.com>

doxygen_to_devhelp: Produce Devhelp index format v2

The old version-1 format is deprecated and now produces warnings.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=106186
Reviewed-by: Philip Withnall <withnall@endlessm.com>
[smcv: Add the .devhelp2 file to .gitignore as suggested]
Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit fa922639207de16d459983144ddad28b7abe60c2)

doxygen_to_devhelp: Make the API reference the front page

The tutorial is not necessarily a great entry point for the libdbus
documentation: it's infrequently updated, and we should probably have
the "If you use this low-level API directly, you're signing up for some
pain" message from the API reference show up in devhelp more immediately.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=106186
Reviewed-by: Philip Withnall <withnall@endlessm.com>
[smcv: Add longer commit message with rationale]
Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit c84ac8b1ef3a9fa18127b2462ead369c4e8846bd)

doc: Install highlight.pack.js if present

Newer versions of yelp-build use this instead of a jQuery syntax
highlighter.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=106171
Reviewed-by: Philip Withnall <withnall@endlessm.com>
[smcv: Also add it to .gitignore as suggested]
Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit 49ad5b110fd5f5f4e41405d98007a11d8eb741f7)

doc: Only install ancillary files from yelp-build if they exist

Newer versions of yelp-build don't install jquery.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=106171
Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
(cherry picked from commit bab857fb6f75ffe0ac3771de4b8272ad97623a2c)

1.12.6

Signed-off-by: Simon McVittie <smcv@collabora.com>

Clarify NEWS entry with implications of fd.o#105165

Signed-off-by: Simon McVittie <smcv@collabora.com>

Update NEWS for #105165

Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit c62a20624e4616e7fa84427c54f3653312a7edb5)

Add a unit test for the dbus-daemon resetting its fd limit

Reviewed-by: David King <dking@redhat.com>
[smcv: Fix typo in cmake macro name]
Signed-off-by: Simon McVittie <smcv@collabora.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=105165
(cherry picked from commit 49ca421997d91d3e01626b2c92a826e6a5db0b2f)

cmake: Check for getrlimit, setrlimit

This gives us feature parity with the Autotools build system for this
particular area, and in particular means a system dbus-daemon built
with cmake can expand its fd limit.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=105165
(cherry picked from commit a146724f2f7610bc0a968d03a3f20481c03a6a37)

bus: raise fd limits before dropping privs

Startup ordering was changed in #92832 to ensure that SELinux audit
messages could be sent. As a side effect, the raising of file descriptor
limits was moved to after the dropping of root privileges, resulting in
the limit change always failing.

Move the raise_file_descriptor_limit() call to ensure that it is called
before dropping root privileges.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=105165
Bug-RedHat: https://bugzilla.redhat.com/show_bug.cgi?id=1529044
[smcv: Call raise_file_descriptor_limit() even if !context->user]
Reviewed-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit 6e42964f5f850f4108fd8f7f3cd385ab4d60f9f6)

Start towards 1.12.6

Signed-off-by: Simon McVittie <smcv@collabora.com>

1.12.4

Signed-off-by: Simon McVittie <smcv@collabora.com>

Add NEWS for #104925

Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit 93433970e9c5a5d0b7f92a37174b40624c711475)

doc: Fix bracket escaping in Ducktype API design file

There’s no need to escape closing brackets if the paired opening bracket
is escaped (or doesn’t need escaping).

See
https://github.com/projectmallard/mallard-ducktype/issues/16#issuecomment-362590519.

Signed-off-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=104925
Reviewed-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit ad72d6bc5f45e78d21f64099f670a89dcf32507d)

Add NEWS for #102839

Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit 705db4455d482edba52d0af3fe57521da9d33e86)

Add new test for waiting on pending calls in threads

Based on code contributed by Manish Narang. This is not included in the
automated test suite, because it isn't reliable on heavily-loaded
automatic test infrastructure like Travis-CI.

Reviewed-by: Philip Withnall <withnall@endlessm.com>
[smcv: Add the test to the CMake build system too, as requested]
[smcv: Convert into a manual test]
Signed-off-by: Simon McVittie <smcv@collabora.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=102839
(cherry picked from commit 0b1e2928606e0cf138c38e75eb248d2ed19bff15)

Add a simplified backport of g_steal_pointer()

This will be used in tests later in the branch.

Sadly we can't use GLIB_VERSION_2_44 unless we are willing to have a
hard dependency on GLib 2.44, which would force us to do all our
Travis-CI builds in Docker containers rather than in ye olde base
system, and that adds 50% to the time taken to do builds.

Reviewed-by: Philip Withnall <withnall@endlessm.com>
[smcv: Rebase onto 1.13.x branch, fix minor conflicts]
Signed-off-by: Simon McVittie <smcv@collabora.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=101354
(cherry picked from commit d5742550ca566317eaddea0ff7db04098f9f044f)

cmake: Match AC_DEFINE more precisely, respecting [] quoting

The regular expression previously used here to select the second
comma-delimited argument won't work when we introduce an argument
containing a comma, which I need to do now. We can address this by
recognising Autoconf's quoting mechanism (which uses square
brackets).

This is not 100% right (it doesn't understand nested square brackets),
but it's good enough in practice.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Acked-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=101354
(cherry picked from commit 83b439f7b4c6a946e0fe9a0287910ba4f6318143)

tests: Add the ability to multiply up test timeouts

Tests that brute-force OOM code paths can be rather slow.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=100317
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit 5c91d85f3ed462dac1e011aab216c9021e826773)

test_connect_to_bus: Allow skipping the use of a DBusLoop

DBusLoop isn't thread-safe, so we can't use it to test multi-threaded
situations.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=102839
Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
(cherry picked from commit f127c8e110faed76039f96dbc53a87f093fea312)
[smcv: Adjust for older codebase]

DBusPendingCall: Improve doc-comments around completed flag

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=102839
Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
(cherry picked from commit 57a0cf1d14c20765bfc7a36234955b14f3811f2a)

DBusPendingCall: Only update ->completed under the connection lock

If one thread is blocking on a pending call, and another thread is
dispatching the connection, then we need them to agree on the value
of the completed flag by protecting all accesses with a lock. Reads
for this member seem to have the connection lock already, so it's
sufficient to make sure that the only write also happens under the
connection lock.

We already set the completed flag before calling the callback, so it
seems OK to stretch it to meaning that some thread has merely *taken
responsibility for* calling the callback.

The completed flag shares a bitfield with timeout_added, but that
flag is protected by the connection lock already.

Based on suggestions from Simon McVittie on
<https://bugs.freedesktop.org/show_bug.cgi?id=102839>.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=102839
[smcv: Revert indentation changes; add commit message]
Reviewed-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit d3e03eb50eefa5a38d87f274c7de73f36468459c)

DBusConnection: Pass a pending call around more often

If a pending call is provided, _dbus_connection_do_iteration_unlocked
checks whether it has completed or has a reply ready as soon as it
acquires the I/O path. If that's the case, then the iteration
terminates without trying to carry out I/O, so that the pending call
can be dispatched immediately, without blocking until a timeout is
reached. This change is believed to be necessary, but not sufficient,
to resolve #102839.

Based on part of a patch from Michael Searle on
<https://bugs.freedesktop.org/show_bug.cgi?id=102839>.
Commit message added by Simon McVittie.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=102839
Reviewed-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit 30f8a38b3c8f8756744d6b65dd8207302a683acc)

NEWS: Mention systemd < 237 here too

Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit 38dea203a5514dc4b4e858fce0b6957d4cf116ab)

tmpfiles: Add a note that one line is not needed with newer systemd

Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit 8fd30736917e36b5eb69167dc3ef23be74f76039)

travis-ci: Get autoconf-archive from Debian 9 'stretch'

Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit acb775a92b1c554f251de129c7e51db19522b65e)

Update NEWS for #104577

Signed-off-by: Simon McVittie <smcv@collabora.com>

Modify systemd tmpfiles.d snippet to create /var/lib/dbus/

This snippet was already attempting to create /var/lib/dbus/machine-id,
but would fail on volatile or stateless systems where /var/lib/dbus/
did not already exist. systemd-tmpfiles automatically creates parent
directories for tmpfiles of type 'd', 'D', etc., but not for files
or symlinks (https://github.com/systemd/systemd/issues/7853).

Signed-off-by: Chris Lesiak <chris.lesiak@licor.com>
[smcv: Extended commit message to clarify why we need this]
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=104577
Reviewed-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit aeebf801f1ab462cba3a174440c09a86d326002a)

Update NEWS for #104265

Signed-off-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit 67d04ad1aa5160fdd36ba2108e8f988ddcffe423)

Resolve bindir variable in dbus-1.pc file correctly

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=104265
Reviewed-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit d5c2e5539fcaf37b9150d2391bcf9d6737931464)

Merge remote-tracking branch 'github/dbus-1.12' into dbus-1.12

_dbus_server_new_for_socket: Iterate over arrays as intended

Commit 0c03b505 was meant to clear all the fds indexed by j in
[0, n_fds), which socket_disconnect() can't be allowed to close
(because on failure the caller remains responsible for closing them);
but instead it closed the one we failed to add to the main loop
(fd i), repeatedly.

Similarly, it was meant to invalidate all the watches indexed by j
in [i, n_fds) (the one we failed to add to the main loop and the ones
we didn't try to add to the main loop yet), which socket_disconnect()
can't be allowed to see (because it would fail to remove them from
the main loop and hit an assertion failure); but instead it invalidated
fd i, repeatedly.

These happen to be the same thing if you only have one fd, resulting
in the test-case passing on an IPv4-only system, but failing on a
system with both IPv4 and IPv6.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=89104
Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
(cherry picked from commit c9aa00ce730f9741ab39ff704e46ec33dd4a11ea)

NEWS for 1.12.x

Signed-off-by: Simon McVittie <smcv@collabora.com>

Add a targeted test for OOM during _dbus_server_new_for_tcp_socket()

This also covers _dbus_server_new_for_socket(), which is one of the
worse places in terms of complexity of the error-unwinding path
(3 labels).

Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=89104
(cherry picked from commit ac6c4d384d83c2e39a4f4156ef11cadcfbfa44f6)

_dbus_listen_tcp_socket: Don't rely on dbus_realloc setting errno

dbus_realloc() doesn't guarantee to set errno (if it did, the
only reasonable thing it could set it to would be ENOMEM). In
particular, faking OOM conditions doesn't set it. This can cause an
assertion failure when OOM tests assert that the only error that can
validly occur is DBUS_ERROR_NO_MEMORY.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=89104
(cherry picked from commit 9ded6907e66b89c3c74620a4485e8f752f092a60)

_dbus_server_new_for_socket: Properly disconnect during error unwinding

_dbus_server_finalize_base() asserts that the socket has been
disconnected, but in some OOM code paths we would call it without
officially disconnecting. Do so.

This means we need to be a bit more careful about what is
socket_disconnect()'s responsibility to clean up, what is
_dbus_server_new_for_socket()'s responsibility, and what is the caller's
responsibility.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=89104
(cherry picked from commit 0c03b505a9718c6d497caffb7d6083371679a852)

_dbus_server_new_for_socket: Invalidate watches during error unwinding

We assert that every watch is invalidated before it is freed, but
in some OOM code paths this didn't happen.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=89104
(cherry picked from commit 1ce34beef85a7a0b3c25890837e3a72f8bdac1f0)

Update NEWS for 1.12.x

Signed-off-by: Simon McVittie <smcv@collabora.com>

Add a missing space in logging output

Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=103729
(cherry picked from commit c9ce0a705966c78c680b465fb5747a16c570c4a8)

Start 1.12.4

Signed-off-by: Simon McVittie <smcv@collabora.com>

1.12.2

Signed-off-by: Simon McVittie <smcv@collabora.com>

Update NEWS for 1.12.x

Signed-off-by: Simon McVittie <smcv@collabora.com>

bus_connections_setup_connection: If we can't set it up, log why

While reviewing fd.o#101354, Philip Withnall pointed out that if we
rejected a connection in the new code there, we didn't log why. It
turns out we didn't log that in the more normal code path either.
Redo the error handling so that failure to set up a connection
is logged.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=103592
(cherry picked from commit 56847ae818e8c144b9c7a8102efd1c926ead2f62)

Update NEWS for 1.12.x

Signed-off-by: Simon McVittie <smcv@collabora.com>

travis-ci: Enable/disable more features in various builds

In the debug build, enable features that are off by default. In the
reduced build, explicitly disable features, some of which are
on by default. In the legacy build, check that we can compile the
default feature-set without inotify, dnotify, systemd, etc.

Reviewed-by: Philip Withnall <withnall@endlessm.com>
[smcv: Rebase onto 1.13.x branch, fix minor conflicts]
Signed-off-by: Simon McVittie <smcv@collabora.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=101354
(cherry picked from commit 3c031ef5aa1f7f53c6344781cb38b78abe44dc96)

_dbus_accept_with_noncefile: Don't leak nonce

This was always leaked, both on success and on error.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=103597
(cherry picked from commit 37d5af203c0fc22c9ae746b2ae4781ff648a792f)

do_noncefile_create: Avoid freeing uninitialized memory on error

We could free all of these without having ever successfully
initialized them.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=103597
(cherry picked from commit 6d08f5c04e601f16ef4ee2126a90c924b7e26df0)

do_check_nonce: Don't free uninitialized memory on OOM

If _dbus_string_init() fails, it doesn't guarantee that the string
is initialized to anything in particular. Worse, if
_dbus_string_init (&buffer) fails, p would never have been initialized
at all, due to the use of the short-circuiting || operator.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=103597
(cherry picked from commit 0ea0e4b0fddd1109835b8b9f7a8319d59c8d9303)

Update NEWS

Signed-off-by: Simon McVittie <smcv@collabora.com>

build: Avoid using `echo -e`

`echo -e` is a GNU extension, and in particular not available on
Darwin.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=103493
[smcv: Added commit message]
Reviewed-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit 216ae1adfcf9ce9274019b3c29b8f1245c26dec1)

Update NEWS

Signed-off-by: Simon McVittie <smcv@collabora.com>

build: Use $(MKDIR_P) syntax suitable for install-sh

This fixes the build when a GNU-compatible `mkdir -p` is not available.

Bug: https://bugs.freedesktop.org/show_bug.cgi?id=103521
[smcv: Add commit message]
Reviewed-by: Simon McVittie <smcv@collabora.com>
(cherry picked from commit c7d8b723eb53848f9a86f36d640ea5f564d34590)

Bump version for 1.12.x branch

Signed-off-by: Simon McVittie <smcv@collabora.com>

Prepare 1.12.0

Signed-off-by: Simon McVittie <smcv@collabora.com>

Disable deprecation warnings for stable branch

We're not going to replace deprecated functions here, similar to commit
88e0ccb2 in the dbus-1.10 branch.

Signed-off-by: Simon McVittie <smcv@collabora.com>

NEWS for #103420

Signed-off-by: Simon McVittie <smcv@collabora.com>

build: Remove various unused files from build system

These were in git but not distributed in source tarballs, and in fact
not hooked up to the Autotools build system at all.
test/data/valid-introspection-files was mentioned in the CMake build
system (copied from the source directory to the build directory), but
according to `git grep` is not used for anything.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=103420

build: Include README.cmake in Autotools "make dist"

Our official source releases are Autotools "make dist" tarballs, but
there's no reason why CMake users can't use those too, and we already
include the CMake build files.

Signed-off-by: Simon McVittie <smcv@collabora.com>
Reviewed-by: Philip Withnall <withnall@endlessm.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=103420