When faced with a `GType` which is present in multiple typelibs, the old
implementation was not deterministic, as it iterated over a hash table
of typelibs. The iteration order of a hash table is not deterministic.
Use the new `ordered_typelibs` and `ordered_lazy_typelibs` arrays to
iterate instead, making the order deterministic.
Add a unit test to check this. In particular, to check that symbols
which are present in both `Gio` and `GioUnix` are correctly resolved as
being from `GioUnix`.
Signed-off-by: Philip Withnall <pwithnall@gnome.org>
Helps: #3303
The iteration code used `g_string_overwrite_len()` to try and simplify
buffer allocation and growth, but seemingly forgot to handle the fact
that it doesn’t nul-terminate what it overwrites: the method is intended
to be used to splice bits into longer strings, not to overwrite an
entire nul-terminated string.
This meant that when iterating over a comma-separated `c_prefix` like
`GUnix,G`, on the second iteration `g_string_overwrite_len()` would be
used to write `G` into index 0 of the already-set `GUnix` string in the
buffer, leading to the first iteration happening all over again and the
`G` prefix being ignored.
This led to symbols failing to be matched to the `GioUnix` typelib, even
though they should have been.
This will be checked by a test in the following commit.
Signed-off-by: Philip Withnall <pwithnall@gnome.org>
Helps: #3303
There are various places where the set of typelibs is iterated over or
returned in an ordered way. In order to keep results deterministic and
reproducible, we need to keep this set ordered.
Keep a `GPtrArray` of the typelibs (one for fully-loaded ones and one
for lazy ones) alongside the existing hash tables. This will be used for
iteration in the next few commits.
Signed-off-by: Philip Withnall <pwithnall@gnome.org>
Helps: #3303
Almost identically to the previous commit, fix a similar latent bug in
`g_dbus_connection_export_action_group()`, which was not ready to handle
the fledgling `GActionGroupExporter` being freed early on an error
handling path.
See the previous commit message for details of the approach.
This includes a unit test.
Signed-off-by: Philip Withnall <pwithnall@gnome.org>
Fixes: #3366
This latent bug wasn’t triggered until commit 3f30ec86c (or its
cherry-pick onto `glib-2-80`, 747e3af99, which was first released in
2.80.1).
That change means that `g_menu_exporter_free()` is now called on the
registration failure path by `g_dbus_connection_register_object()`
before it returns. The caller then tries to call `g_slice_free()` on the
exporter again. The call to `g_menu_exporter_free()` tries to
dereference/free members of the exporter which it expects to be
initialised — but because this is happening in an error handling path,
they are not initialised.
If it were to get any further, the `g_slice_free()` would then be a
double-free on the exporter allocation.
Fix that by making `g_menu_exporter_free()` robust to some of the
exporter members being `NULL`, and moving some of the initialisation
code higher in `g_dbus_connection_export_menu_model()`, and removing the
duplicate free code on the error handling path.
This includes a unit test.
Signed-off-by: Philip Withnall <pwithnall@gnome.org>
Fixes: #3366
In glocalfile we're allocating some temporary strings but we don't free
them on early returns, so free them once done and unset the variables
to prevent them being used incorrectly.
The type used when declaring a bitfield member of a struct doesn't
affect the amount of space allocated for it - only whether it's signed
or unsigned. In standard C99 (6.2.7.1), only _Bool, signed int and
unsigned int or typedefs to them are allowed as bitfield types, but GCC
allows other integer types as an extension.
In this case, the GIBaseInfo and GIBaseInfoStack structs are meant to
have identical layout. However, type_is_embedded was declared as an
unsigned bitfield in the former and a uint32_t in the latter. This was
harmless on most platforms because the following member is an aligned
pointer, but (for example) on m68k-linux-gnu pointers only need to be
16-bit aligned, so GCC only allocates 16 bits for the bitfield.
Change the type in the declaration to unsigned int, and add an padding
bitfield following it to ensure there's space for 32 bits on all
platforms in the future.
Signed-off-by: Adam Sampson <ats@offog.org>
GCC 14 now emits this warning with the tests:
```
In file included from ../glib/gthread.h:34,
from ../glib/gasyncqueue.h:34,
from ../glib/glib.h:34,
from ../glib/tests/atomic.c:14:
../glib/tests/atomic.c: In function 'test_types':
../glib/gatomic.h:140:5: error: argument 2 of '__atomic_store' discards 'volatile' qualifier [-Werror=discarded-qualifiers]
140 | __atomic_store (gaps_temp_atomic, &gaps_temp_newval, __ATOMIC_SEQ_CST); \
| ^~~~~~~~~~~~~~
../glib/tests/atomic.c:139:3: note: in expansion of macro 'g_atomic_pointer_set'
139 | g_atomic_pointer_set (&vp_str_vol, NULL);
| ^~~~~~~~~~~~~~~~~~~~
cc1.exe: all warnings being treated as errors
```
I can’t think of a way to cast around this in the definition of
`g_atomic_pointer_set()` without making the behaviour worse (less type
safe) for modern non-volatile atomic variables.
We would like to strongly nudge users of GLib away from declaring atomic
variables as `volatile`, so letting another compiler warning be emitted
when they do is not the end of the world. As long as it doesn’t stop old
code compiling (without `-Werror`).
Signed-off-by: Philip Withnall <pwithnall@gnome.org>
Tests may have runtime dependencies that are related to the typelib
dependencies, so we need to satify them or the tests will fail at
runtime if we're not building their prerequisite for other reasons.
That's saying that the tests are currently failing when explicitly
running as standalone in meson.
Co-Authored-By: Philip Withnall <philip@tecnocode.co.uk>
We're now caching arg0 but such value is not cleared when a new body is
set as it's in the connection filter test cases where we've a leak as
highlighted by both valgrind and leak sanitizer
In a D-Bus-Specification-compliant message bus, the owner of a well-known
name is a unique name. However, ibus has its own small implementation
of a message bus (src/ibusbus.c) in which org.freedesktop.IBus is
special-cased to also have itself as its owner (like org.freedesktop.DBus
on a standard message bus), and connects to that bus with the
G_DBUS_CONNECTION_FLAGS_MESSAGE_BUS_CONNECTION flag. The ability to do
this regressed when CVE-2024-34397 was fixed.
Relax the checks to allow the owner of a well-known name to be any valid
D-Bus name, even if it is not syntactically a unique name.
Fixes: 683b14b9 "gdbus: Track name owners for signal subscriptions"
Resolves: https://gitlab.gnome.org/GNOME/glib/-/issues/3353
Bug-Debian: https://bugs.debian.org/1070730
Bug-Debian: https://bugs.debian.org/1070736
Bug-Debian: https://bugs.debian.org/1070743
Bug-Debian: https://bugs.debian.org/1070745
Signed-off-by: Simon McVittie <smcv@debian.org>
548ec9f186 accidentally moved the GVariant
spec to the toplevel /usr/share/doc directory, which is surely not
right. Let's move it back into the glib-2.0 subdirectory.
It's debatable whether this is the best place to install the GVariant
specification, since it's not part of the gi-docgen docs, but surely
it's much better than not putting it in any subdirectory.
Fixes#3351
The test case assumes signals will dispatched in a different order than
they're subscribed. In fact, signals can be dispatched in any order,
and are often dispatched in order.
This commit reorders the subscriptions so they're in order, which is
more logical, and also changes the code to only exit the event loops
when there are no pending handlers ready to dispatch.
(cherry picked from commit 47866d252f)
GDBusConnection sends each signal to recipients in a separate idle
callback, and there's no particular guarantee about the order in which
they're scheduled or dispatched. For the NameOwnerChanged signal that
reports the name becoming unowned, it's possible that g_bus_watch_name()
gets its idle callback called before the GDBusProxy:g-name-owner
machinery has updated the name owner, in which case the assertion
will fail.
Fixing GNOME/glib#3268 introduced a new subscription to NameOwnerChanged
which can alter the order of delivery, particularly in the case where
G_DBUS_PROXY_FLAGS_NO_MATCH_RULE was used (as tested in
/gdbus/proxy/no-match-rule). The resulting test failure is intermittent,
but reliably appears within 100 repetitions of that test.
Fixes: 511c5f5b "tests: Wait for gdbus-testserver to die when killing it"
Signed-off-by: Simon McVittie <smcv@debian.org>
If the file to be added is on a read-only filesystem, opening read/write
will fail with EROFS. In this case we should fall back to opening it
read-only, the same way we already do if write access is forbidden by
DAC or MAC.
An easy way to reproduce this test failure is to build and test GLib
in a podman container, with its source code read-only and its build
directory read/write:
podman run --rm -it \
-v $(pwd):$(pwd):ro \
-v $(pwd)/_build:$(pwd)/_build:rw \
-w $(pwd) ...
Before this commit, the dbus-appinfo test would fail, because opening
${srcdir}/gio/tests/org.gtk.test.dbusappinfo.flatpak.desktop read/write
would fail with EROFS.
For completeness, give similar handling to the other error codes
documented in Linux open(2) that might succeed if re-attempted using
read-only access: according to that documentation, we could get EPERM
if opening read/write is prevented by fcntl F_ADD_SEALS, or ETXTBSY
if the file is an executable that is currently being run.
Signed-off-by: Simon McVittie <smcv@collabora.com>
This was a bug that existed during development of this branch; make sure
it doesn't come back.
This test fails with a use-after-free and crash if we comment out the
part of name_watcher_unref_watched_name() that removes the name watcher
from `map_method_serial_to_name_watcher`.
It would also fail with an assertion failure if we asserted in
name_watcher_unref_watched_name() that get_name_owner_serial == 0
(i.e. that GetNameOwner is not in-flight at destruction).
Signed-off-by: Simon McVittie <smcv@collabora.com>
The vulnerability reported as GNOME/glib#3268 can be characterized
as: these signals from an attacker should not be delivered to either
the GDBusConnection or the GDBusProxy, but in fact they are (in at
least some scenarios).
Reproduces: https://gitlab.gnome.org/GNOME/glib/-/issues/3268
Signed-off-by: Simon McVittie <smcv@collabora.com>
The expected result is that because TEST_CONN_SERVICE owns
ALREADY_OWNED_NAME but not (yet) OWNED_LATER_NAME, the signal will be
delivered to the subscriber for the former but not the latter.
Before #3268 was fixed, it was incorrectly delivered to both.
Reproduces: https://gitlab.gnome.org/GNOME/glib/-/issues/3268 (partially)
Signed-off-by: Simon McVittie <smcv@collabora.com>
Otherwise a malicious connection on a shared bus, especially the system
bus, could trick GDBus clients into processing signals sent by the
malicious connection as though they had come from the real owner of a
well-known service name.
Resolves: https://gitlab.gnome.org/GNOME/glib/-/issues/3268
Signed-off-by: Simon McVittie <smcv@collabora.com>
We will use this in a subsequent commit to prevent signals from an
impostor from being delivered to a subscriber.
To avoid message reordering leading to misleading situations, this does
not use the existing mechanism for watching bus name ownership, which
delivers the ownership changes to other main-contexts. Instead, it all
happens on the single thread used by the GDBusWorker, so the order in
which messages are received is the order in which they are processed.
Signed-off-by: Simon McVittie <smcv@collabora.com>
This will become confusing when we start tracking the owner of a
well-known-name sender, and it's redundant anyway. Instead, track the
1 bit of data that we actually need: whether it's a well-known name.
Strictly speaking this too is redundant, because it's syntactically
derivable from the sender, but only via extra string operations.
A subsequent commit will add a data structure to keep track of the
owner of a well-known-name sender, at which point this boolean will
be replaced by the presence or absence of that data structure.
Signed-off-by: Simon McVittie <smcv@collabora.com>
No functional change, just removing some nesting. The check for whether
signal_data->subscribers is empty changes from a conditional that tests
whether it is into an early-return if it isn't.
A subsequent commit will add additional conditions that make us consider
a SignalData to be still in use and therefore not eligible to be removed.
Signed-off-by: Simon McVittie <smcv@collabora.com>
No functional changes, except that the implicit ownership-transfer
for the rule field becomes explicit (the local variable is set to NULL
afterwards).
Signed-off-by: Simon McVittie <smcv@collabora.com>
