Previously we detected glibc support at compile time and used
it unconditionally; better to try it and fall back, this way
we continue to run on older kernels when compiled for newer ones.
Previously, the watch handler would block until the I/O path was available.
However, if another non-main thread was doing a synchronous call, this would
cause the main thread to block on that thread, a highly undesirable
thing because it's important for the main thread to remain responsive
for user interfaces.
Signed-off-by: Colin Walters <walters@verbum.org> Signed-off-by: Thiago Macieira <thiago@kde.org>
Colin Walters [Thu, 18 Feb 2010 20:33:28 +0000 (15:33 -0500)]
Monitor service directories for changes
It's not expected to have to manually SIGHUP the bus after installing
a new .service file. Since our directory monitoring is already set
up to queue a full reload which includes service activation, simply
monitor the servicedirs too.
Colin Walters [Tue, 2 Feb 2010 19:57:47 +0000 (14:57 -0500)]
Fix compilation in --disable-selinux case
_dbus_change_to_daemon_user moved into selinux.c for the --with-selinux
(and audit) case because that's where all of the relevant libcap headers
were being used. However in the --disable-selinux case this didn't
compile and wasn't very clean.
If we don't have libaudit, use the legacy direct setgid/setuid bits
we had before in dbus-sysdeps-util-unix.c.
Colin Walters [Tue, 2 Feb 2010 17:37:17 +0000 (12:37 -0500)]
Fix inotify shutdown
We were incorrectly passing NULL for a DBusList when the usage expected
is a pointer to a NULL DBusList pointer. Also during dbus_shutdown
we need to actually close the inotify fd, and remove our watch.
Move the shutdown handler out of bus.c and into inotify where we
can do all of this cleanly.
Tom Hughes [Sun, 13 Dec 2009 21:30:09 +0000 (13:30 -0800)]
Use monotonic clock for _dbus_get_current_time() if it's available.
_dbus_get_current_time() is used for timeouts, but uses gettimeofday(), which
relies on the wall clock time, which can change. If the time is changed forwards
or backwards, the timeouts are no longer valid, so the monotonic clock must be used.
Colin Walters [Thu, 28 Jan 2010 21:26:39 +0000 (16:26 -0500)]
Clean up inotify watch handling
Substantially based on a patch by Matthias Clasen <mclasen@redhat.com>
kqueue implementation by Joe Marcus Clarke <marcus@freebsd.org>
Previously, when we detected a configuration change (which included
the set of config directories to monitor for changes), we would
simply drop all watches, then readd them.
The problem with this is that it introduced a race condition where
we might not be watching one of the config directories for changes.
Rather than dropping and readding, change the OS-dependent monitoring
API to simply take a new set of directories to monitor. Implicit
in this is that the OS-specific layer needs to keep track of the
previously monitored set.
Colin Walters [Thu, 28 Jan 2010 20:04:14 +0000 (15:04 -0500)]
Don't drop pending activations when reloading configuration
The reload handling for activation simply dropped all knowledge
of pending activations, which was clearly wrong. Refactor things
so that reload only reloads directories, server address etc.
Based on a patch originally from Matthias Clasen <mclasen@redhat.com>
Colin Walters [Thu, 28 Jan 2010 00:38:44 +0000 (19:38 -0500)]
Switch to libcap-ng, avoid linking libdbus against libcap[-ng]
(Commit message written by Colin Walters <walters@verbum.org>)
A current Fedora goal is to convert projects to libcap-ng which
more easily allows dropping Linux capabilities. For software
which also links to libdbus, it's problematic to link against
libcap as well.
Though really, libdbus should have never linked against libcap
in the first place, which is another thing this patch changes
by moving the libcap-using bits out of dbus/ and into bus/.
Colin Walters [Mon, 14 Dec 2009 23:12:24 +0000 (18:12 -0500)]
Ignore exit code zero from activated services
A variety of system components have migrated from legacy init into DBus
service activation. Many of these system components "daemonize", which
involves forking. The DBus activation system treated an exit as an
activation failure, assuming that the child process which grabbed the
DBus name didn't run first.
While we're in here, also differentiate in this code path between the
servicehelper (system) versus direct activation (session) paths. In
the session activation path our error message mentioned a helper
process which was confusing, since none was involved.
Based on a patch and debugging research from Ray Strode <rstrode@redhat.com>
Will Thompson [Wed, 28 Oct 2009 19:40:38 +0000 (19:40 +0000)]
Print all-printable-ASCII byte arrays as strings
In practice, ay seems to be used mostly for binary data (in which case,
hex output is fine) or for Unix file paths (because they may be
non-UTF-8) and similar human-readable strings. So let's print the latter
similarly to strings.
Will Thompson [Wed, 9 Sep 2009 19:35:13 +0000 (20:35 +0100)]
Make array-printing code easier to follow
Previously dbus_message_iter_get_arg_type() was called twice: once in
the loop condition to update 'current_type', and once to check if the
loop will run again. This patch moves updating current_type to the end
of the loop body.
James Westby [Thu, 1 Oct 2009 14:09:54 +0000 (15:09 +0100)]
Correct timeout handling
The timeout handling code subtracts the elapsed time from the timeout
each time a message is received, which drastically reduces the timeout
in circumstances such as service activation.
Correct so that the timeout is never modified, and the elapsed time
instead subtracted where necessary.
Signed-off-by: James Westby <jw+debian@jameswestby.net> Signed-off-by: Scott James Remnant <scott@ubuntu.com>
Colin Watson [Fri, 18 Sep 2009 17:23:39 +0000 (18:23 +0100)]
Fix link order: system libraries should come after libdbus-convenience
libdbus-convenience may use system libraries, but not the other way
round. Most platforms don't care, but on some platforms this means that
system libraries need to be listed after libdbus-convenience.la on the
link line.
dbus-monitor: use unbuffered stdout instead of handling SIGINT
The current SIGINT handling of dbus-monitor ain't making too many people
happy since it defers the exit to the next msg received -- which might
be quite some time away often enough.
This patch replaces the SIGINT handling by simply enabling line-buffered
IO for STDOUT so that even if you redirect dbus-monitor into a file no
lines get accidently lost and the effect of C-c is still immediate.
halfline came up with the great idea to use setvbuf here instead of
fflush()ing after each printf().
(Oh and the old signal handler was broken anyway, the flag should have
been of type sigatomic_t and be marked volatile)
Colin Walters [Tue, 14 Jul 2009 15:43:54 +0000 (11:43 -0400)]
Cope with dbus-launch not being in DBUS_BINDIR
This is a temporary hack for systems which use DBUS_BINDIR=/bin,
but then move dbus-launch back into /usr/bin. Longer term,
we should explicitly support this in upstream code, or even better
figure out how to move dbus-launch into /bin (e.g. dynamically
load libX11 if available), or have a --with-x11-tools configure
option.
(cherry picked from commit 70c5285eb4b757fdd4e552e002c5d54072492152)
Colin Walters [Mon, 13 Jul 2009 17:11:20 +0000 (13:11 -0400)]
Bug 22415 - COYPING file should note we allow later versions of GPL
The canonical copyright information in the source file says we
allow later versions of the GPL. So note that in COPYING too.
(cherry picked from commit f908daed82d61df44710a8ec681558c6d23b940d)
Colin Walters [Sat, 11 Jul 2009 01:33:02 +0000 (21:33 -0400)]
Bug 896 - Avoid race conditions reading message from exited process
Patch based on extensive work from Michael Meeks <michael.meeks@novell.com>,
thanks to Dafydd Harries <dafydd.harries@collabora.co.uk>,
Kimmo Hämäläinen <kimmo.hamalainen@nokia.com> and others.
The basic idea with this bug is that we effectively ignore errors
on write. Only when we're done reading from a connection do we
close down a connection. This avoids a race condition where
if a process (such as dbus-send) exited while we still had
data to read in the buffer, we'd miss that data.
(cherry picked from commit 0e36cdd54964c4012acec2bb8e598b85e82d2846)
Colin Walters [Mon, 13 Jul 2009 17:02:21 +0000 (13:02 -0400)]
Bug 14259 - Refactor _dbus_get_autolaunch_address
Split out the process-launching code, which can be reused for
other applications; in particular, a forthcoming patch to parse
output from launchd for MacOS X.
(cherry picked from commit 6b163e95e7a2318a98c16c0d0944337e38e62efa)
Colin Walters [Mon, 13 Jul 2009 16:47:19 +0000 (12:47 -0400)]
Bug 14259 - Make session address lookup system-dependent
On some platforms such as MacOS X and Windows, we can't depend
on an environment variable to determine the address of the
session bus. Create a sysdep function dbus_lookup_session_address
which can be filled in with platform-specific code.
(cherry picked from commit 6478ec6949c6bb794237b43d03b68f80eba1288c)
Colin Walters [Sat, 11 Jul 2009 02:27:55 +0000 (22:27 -0400)]
Bug 18121 - Use a monotonic clock for pthread timeouts
Patch based on one from Keith Mok <ek9852@gmail.com>, some
followup work from Janne Karhunen <Janne.Karhunen@gmail.com>.
We don't want condition variable timeouts to be affected by the system clock.
Use the POSIX CLOCK_MONOTONIC if available.
(cherry picked from commit ae24bb35e2ee3ecde990f55852982b573754ec43)
Will Thompson [Sun, 7 Jun 2009 16:44:26 +0000 (17:44 +0100)]
Ensure messages are locked while marshalling.
Locking a message has the side-effect of updating the message's length
header. Previously, if dbus_message_marshal() was called on an unlocked
message, it could yield an invalid message (as discovered by Ben
Schwartz in <http://bugs.freedesktop.org/show_bug.cgi?id=19723>).
(cherry picked from commit 9f825271f9106c23fe51ab54abdb5156b7751014)
Add test case for assert when unwinding a container.
* dbus/dbus-message-util.c: when constructing an array of structures,
it's not possible to unwind in case of an error part-way through.
This test will therefore assert.
* test/test-service.c (handle_delay_echo, path_message_func): Add a
variant of the Echo method which sleeps for a short time.
* test/name-test/test-pending-call-timeout.c: Run tests with default,
specified and infinite timeout to make sure we get the reply.
* test/name-test/run-test.sh: Run the new test
* test/name-test/Makefile.am: Build the new test
Expire list timeout may be negative for no expiry.
* bus/expirelist.c (do_expiration_with_current_time): Don't check for
expiry if expire_after is negative, will just disable the expiry timer
after the call.
* dbus/dbus-connection.c (_dbus_connection_block_pending_call): Rework
the timeout math so instead of calculating an end time, which may
overflow, we instead calculate the elapsed time which is always
smaller than the boundaries.
Update documentation now that INT_MAX means no timeout.
* dbus/dbus-connection.c (dbus_connection_send_with_reply): Fix
documentation now that INT_MAX will not be clamped.
(dbus_connection_send_with_reply_and_block): Update documentation too.
Don't allocate DBusTimeout for pending call when passed INT_MAX
* dbus/dbus-pending-call.c (_dbus_pending_call_new_unlocked): When passed
INT_MAX, do not clamp the value and do not allocate a timeout for the call
(_dbus_pending_call_get_timeout_unlocked): Document that this may return
NULL.
* dbus/dbus-connection.c (_dbus_connection_block_pending_call): Allow the
pending call to have no timeout, in which case we simply block until we
complete, have data or get disconnected.
* dbus/dbus-connection.c (_dbus_connection_attach_pending_call_unlocked):
Don't assume that the pending call has a timeout.
(connection_timeout_and_complete_all_pending_call_unlocked): check that
the timeout was actually added before removing it; this safeguards us
if the pending call doesn't have a timeout.
Marc Mutz [Wed, 7 Jan 2009 11:46:53 +0000 (12:46 +0100)]
configure.in: fail abstract socket test gracefully when cross-compiling
* configure.in: only run AC_CACHE_CHECK if enable_abstract_sockets=auto
* configure.in: warn that, when cross-compiling, we're unable to detect
abstract sockets availability automatically
libselinux behavior in permissive mode wrt invalid domains
Stephen Smalley wrote:
> On Tue, 2009-04-21 at 16:32 -0400, Joshua Brindle wrote:
>
>> Stephen Smalley wrote:
>>
>>> On Thu, 2009-04-16 at 20:47 -0400, Eamon Walsh wrote:
>>>
>>>> Stephen Smalley wrote:
>>>>
>> <snip>
>>
>>
>>> No, I don't want to change the behavior upon context_to_sid calls in
>>> general, as we otherwise lose all context validity checking in
>>> permissive mode.
>>>
>>> I think I'd rather change compute_sid behavior to preclude the situation
>>> from arising in the first place, possibly altering the behavior in
>>> permissive mode upon an invalid context to fall back on the ssid
>>> (process) or the tsid (object). But I'm not entirely convinced any
>>> change is required here.
>>>
>>>
>> I just want to follow up to make sure we are all on the same page here. Was the
>> suggestion to change avc_has_perm in libselinux or context_to_sid in the kernel
>> or leave the code as is and fix the callers of avc_has_perm to correctly handle
>> error codes?
>>
>> I prefer the last approach because of Eamon's explanation, EINVAL is already
>> passed in errno to specify the context was invalid (and if object managers
>> aren't handling that correctly now there is a good chance they aren't handling
>> the ENOMEM case either).
>>
>
> I'd be inclined to change compute_sid (not context_to_sid) in the kernel
> to prevent invalid contexts from being formed even in permissive mode
> (scenario is a type transition where role is not authorized for the new
> type). That was originally to allow the system to boot in permissive
> mode. But an alternative would be to just stay in the caller's context
> (ssid) in that situation.
>
> Changing the callers of avc_has_perm() to handle EINVAL and/or ENOMEM
> may make sense, but that logic should not depend on enforcing vs.
> permissive mode.
>
>
FWIW, the following patch to D-Bus should help:
bfo21072 - Log SELinux denials better by checking errno for the cause
Note that this does not fully address the bug report since
EINVAL can still be returned in permissive mode. However the log
messages will now reflect the proper cause of the denial.
Signed-off-by: Eamon Walsh <ewalsh@tycho.nsa.gov> Signed-off-by: Colin Walters <walters@verbum.org>
William Lachance [Tue, 21 Apr 2009 17:51:46 +0000 (13:51 -0400)]
Bug 19567 - Make marshaling code usable without DBusConnection
Some projects want to reuse the DBus message format, without
actually going through a DBusConnection. This set of changes
makes a few functions from DBusMessage public, and adds a new
function to determine the number of bytes needed to demarshal
a message.
Colin Walters [Wed, 25 Feb 2009 16:10:15 +0000 (11:10 -0500)]
Bug 20137 - Fix alignment usage when demarshaling basics
We can't safely type-pun from e.g. char * to DBusBasicValue *, because
the latter has higher alignment requirements. Instead, create an
explicit pointer for each case.
Also, we mark each one volatile to sidestep strict aliasing issues, for
the future when we turn on strict aliasing support.
Original patch and review from Jay Estabrook <jay.estabrook@hp.com>.
Colin Walters [Wed, 1 Apr 2009 16:02:00 +0000 (12:02 -0400)]
Bug 17803 - Fix both test case and validation logic
The previous commit had errors in both the test case and
the validation logic. The test case was missing a trailing
comma before the previous one, so we weren't testing the
signature we thought we were.
The validation logic was wrong because if the type was not valid,
we'd drop through the entire if clause, and thus skip returning
an error code, and accept the signature.
projects / thirdparty / dbus.git / log
