X11 clients tend to assume that pointers have buttons. This
assumption means they often fail to handle the X error that
is generated when querying the button mapping of a pointer
device that lacks buttons.
This failure to handle the X error leads to those client
applications to abruptly exit.
This commit assigns vestigial buttons to the gesture pointer
device for the sole purpose of backward compatibility with
legacy X11 clients.
That technique is already employed for a different pointer,
the relative pointer device, for similar reasons, so this
just makes the legacy client compatibility more complete.
See https://gitlab.gnome.org/GNOME/mutter/-/issues/2353
(cherry picked from commit 456b0e86bb)
The cursor in DIX is actually split in two parts, the cursor itself and
the cursor bits, each with their own devPrivates.
The cursor itself includes the cursor bits, meaning that the cursor bits
devPrivates in within structure of the cursor.
Both Xephyr and Xwayland were using the private key for the cursor bits
to store the data for the cursor, and when using XSELINUX which comes
with its own special devPrivates, the data stored in that cursor bits'
devPrivates would interfere with the XSELINUX devPrivates data and the
SELINUX security ID would point to some other unrelated data, causing a
crash in the XSELINUX code when trying to (re)use the security ID.
CVE-2024-0409
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
(cherry picked from commit 2ef0f1116c)
The XSELINUX code will label resources at creation by checking the
access mode. When the access mode is DixCreateAccess, it will call the
function to label the new resource SELinuxLabelResource().
However, GLX buffers do not go through the XACE hooks when created,
hence leaving the resource actually unlabeled.
When, later, the client tries to create another resource using that
drawable (like a GC for example), the XSELINUX code would try to use
the security ID of that object which has never been labeled, get a NULL
pointer and crash when checking whether the requested permissions are
granted for subject security ID.
To avoid the issue, make sure to call the XACE hooks when creating the
GLX buffers.
Credit goes to Donn Seeley <donn@xmission.com> for providing the patch.
CVE-2024-0408
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Acked-by: Peter Hutterer <peter.hutterer@who-t.net>
(cherry picked from commit e5e8586a12)
Disabling a master device floats all slave devices but we didn't do this
to already-disabled slave devices. As a result those devices kept their
reference to the master device resulting in access to already freed
memory if the master device was removed before the corresponding slave
device.
And to match this behavior, also forcibly reset that pointer during
CloseDownDevices().
Related to CVE-2024-21886, ZDI-CAN-22840
(cherry picked from commit 26769aa71f)
The `DisableDevice()` function is called whenever an enabled device
is disabled and it moves the device from the `inputInfo.devices` linked
list to the `inputInfo.off_devices` linked list.
However, its link/unlink operation has an issue during the recursive
call to `DisableDevice()` due to the `prev` pointer pointing to a
removed device.
This issue leads to a length mismatch between the total number of
devices and the number of device in the list, leading to a heap
overflow and, possibly, to local privilege escalation.
Simplify the code that checked whether the device passed to
`DisableDevice()` was in `inputInfo.devices` or not and find the
previous device after the recursion.
CVE-2024-21886, ZDI-CAN-22840
This vulnerability was discovered by:
Jan-Niklas Sohn working with Trend Micro Zero Day Initiative
(cherry picked from commit bc1fdbe465)
The `XISendDeviceHierarchyEvent()` function allocates space to store up
to `MAXDEVICES` (256) `xXIHierarchyInfo` structures in `info`.
If a device with a given ID was removed and a new device with the same
ID added both in the same operation, the single device ID will lead to
two info structures being written to `info`.
Since this case can occur for every device ID at once, a total of two
times `MAXDEVICES` info structures might be written to the allocation.
To avoid it, once one add/remove master is processed, send out the
device hierarchy event for the current state and continue. That event
thus only ever has exactly one of either added/removed in it (and
optionally slave attached/detached).
CVE-2024-21885, ZDI-CAN-22744
This vulnerability was discovered by:
Jan-Niklas Sohn working with Trend Micro Zero Day Initiative
(cherry picked from commit 4a5e9b1895)
There's a racy sequence where a master device may copy the button class
from the slave, without ever initializing numButtons. This leads to a
device with zero buttons but a button class which is invalid.
Let's copy the numButtons value from the source - by definition if we
don't have a button class yet we do not have any other slave devices
with more than this number of buttons anyway.
CVE-2024-0229, ZDI-CAN-22678
This vulnerability was discovered by:
Jan-Niklas Sohn working with Trend Micro Zero Day Initiative
(cherry picked from commit df3c65706e)
The previous code only made sense if one considers buttons and keys to
be mutually exclusive on a device. That is not necessarily true, causing
a number of issues.
This function allocates and fills in the number of xEvents we need to
send the device state down the wire. This is split across multiple
32-byte devices including one deviceStateNotify event and optional
deviceKeyStateNotify, deviceButtonStateNotify and (possibly multiple)
deviceValuator events.
The previous behavior would instead compose a sequence
of [state, buttonstate, state, keystate, valuator...]. This is not
protocol correct, and on top of that made the code extremely convoluted.
Fix this by streamlining: add both button and key into the deviceStateNotify
and then append the key state and button state, followed by the
valuators. Finally, the deviceValuator events contain up to 6 valuators
per event but we only ever sent through 3 at a time. Let's double that
troughput.
CVE-2024-0229, ZDI-CAN-22678
This vulnerability was discovered by:
Jan-Niklas Sohn working with Trend Micro Zero Day Initiative
(cherry picked from commit 219c54b8a3)
If a device has both a button class and a key class and numButtons is
zero, we can get an OOB write due to event under-allocation.
This function seems to assume a device has either keys or buttons, not
both. It has two virtually identical code paths, both of which assume
they're applying to the first event in the sequence.
A device with both a key and button class triggered a logic bug - only
one xEvent was allocated but the deviceStateNotify pointer was pushed on
once per type. So effectively this logic code:
int count = 1;
if (button && nbuttons > 32) count++;
if (key && nbuttons > 0) count++;
if (key && nkeys > 32) count++; // this is basically always true
// count is at 2 for our keys + zero button device
ev = alloc(count * sizeof(xEvent));
FixDeviceStateNotify(ev);
if (button)
FixDeviceStateNotify(ev++);
if (key)
FixDeviceStateNotify(ev++); // santa drops into the wrong chimney here
If the device has more than 3 valuators, the OOB is pushed back - we're
off by one so it will happen when the last deviceValuator event is
written instead.
Fix this by allocating the maximum number of events we may allocate.
Note that the current behavior is not protocol-correct anyway, this
patch fixes only the allocation issue.
Note that this issue does not trigger if the device has at least one
button. While the server does not prevent a button class with zero
buttons, it is very unlikely.
CVE-2024-0229, ZDI-CAN-22678
This vulnerability was discovered by:
Jan-Niklas Sohn working with Trend Micro Zero Day Initiative
(cherry picked from commit ece23be888)
Both DeviceFocusEvent and the XIQueryPointer reply contain a bit for
each logical button currently down. Since buttons can be arbitrarily mapped
to anything up to 255 make sure we have enough bits for the maximum mapping.
CVE-2023-6816, ZDI-CAN-22664, ZDI-CAN-22665
This vulnerability was discovered by:
Jan-Niklas Sohn working with Trend Micro Zero Day Initiative
(cherry picked from commit 9e2ecb2af8)
It caused an incorrect result of the blend operation.
Use glColorMask to prevent non-1.0 alpha channel values in a depth 32
pixmap backing an effective depth 24 window. For blending operations,
the expectation is that the destination drawable contains valid pixel
values, so the alpha channel should already be 1.0.
Fixes: d1f142891e ("glamor: Ignore destination alpha as necessary for composite operation")
Issue: https://gitlab.gnome.org/GNOME/mutter/-/issues/3104
(cherry picked from commit d1bbf82d72)
XTest requests lets the client specify a device ID, only if none
is specified do we fall back to the XTEST special device.
As of commit
aa4074251 input: Add new hook DeviceSendEventsProc for XTEST
regular devices are no longer able to send XTest events because they
have no sendEventsProc set.
This caused issue #1574 and the crash was fixed with commit
e820030de xtest: Check whether there is a sendEventsProc to call
but we still cannot send XTest events through a specific device.
Fix this by defaulting every device to the XTest send function and
punting it to the DDX (i.e. Xwayland) to override the devices as
necessary.
Fixes e820030de2
Fixes aa4074251f
(cherry picked from commit de0031eefd)
Otherwise only XTest events on the XTest device get handled, XTest
requests on real devices are still processed as normal events.
(cherry picked from commit 7f7adfdef8)
This just makes the existing behavior explicit, previously we relied on
a malloc(numAxes * ...) to return NULL to error out.
(cherry picked from commit 7aba2514b2)
If we remove a master device and specify which other master devices
attached slaves should be returned to, enforce that those two are
indeeed a pointer and a keyboard.
Otherwise we can try to attach the keyboards to pointers and vice versa,
leading to possible crashes later.
(cherry picked from commit 37539cb0bf)
Xwayland uses OEFFIS_DEVICE_ALL_DEVICES to get all possible device types
enabled.
Be more selective and specify explicitly keyboard and pointer instead of
relying on what "all devices" translates to in the stack.
See-also: https://gitlab.gnome.org/GNOME/mutter/-/issues/3194
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
(cherry picked from commit 1bf4d60acd)
button->xkb_acts is supposed to be an array sufficiently large for all
our buttons, not just a single XkbActions struct. Allocating
insufficient memory here means when we memcpy() later in
XkbSetDeviceInfo we write into memory that wasn't ours to begin with,
leading to the usual security ooopsiedaisies.
CVE-2023-6377, ZDI-CAN-22412, ZDI-CAN-22413
This vulnerability was discovered by:
Jan-Niklas Sohn working with Trend Micro Zero Day Initiative
(cherry picked from commit 0c1a93d319)
Affected are ProcRRChangeProviderProperty and ProcRRChangeOutputProperty.
See also xserver@8f454b79 where this same bug was fixed for the core
protocol and XI.
This fixes an OOB read and the resulting information disclosure.
Length calculation for the request was clipped to a 32-bit integer. With
the correct stuff->nUnits value the expected request size was
truncated, passing the REQUEST_FIXED_SIZE check.
The server then proceeded with reading at least stuff->num_items bytes
(depending on stuff->format) from the request and stuffing whatever it
finds into the property. In the process it would also allocate at least
stuff->nUnits bytes, i.e. 4GB.
CVE-2023-6478, ZDI-CAN-22561
This vulnerability was discovered by:
Jan-Niklas Sohn working with Trend Micro Zero Day Initiative
(cherry picked from commit 14f480010a)
When creating the output with the default "XWAYLAND<n>" name, we use
the MAX_OUTPUT_NAME value to allocate a lot more memory than necessary
to accommodate for future output names once they get updated, but by
doing so, we also send XRandR way too much (zeroed) data since the
"nameLength" value is (purposely) set too big.
So, instead, let's just update the name after creating the RR output,
this way we set both the name and nameLength to their correct values
while keeping the initial large allocation.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Fixes: 3c07a01c42 - xwayland: Use xdg-output name for XRandR
Reviewed-by: Simon Ser <contact@emersion.fr>
(cherry picked from commit 83453fb51e)
At creation, Xwayland uses a generic output name ("XWAYLAND0", etc.) for
the XRandR outputs, and later, once the name is known from the Wayland
protocols, updates the output names using the actual names from the
Wayland compositor.
However, when doing so, it simply updates the string, the "nameLength"
isn't updated, so the name passed to the clients might either end up
being truncated or contain portions of the previous (initial) output
name.
Note, this is using a fixed size buffer initialized with zeros, so this
cannot leak any data other than the previous output name, so this is
mainly a cosmetic issue.
Update the output's "nameLength" when updating the output name.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Fixes: 3c07a01c42 - xwayland: Use xdg-output name for XRandR
Reviewed-by: Simon Ser <contact@emersion.fr>
(cherry picked from commit 0e314afef6)
Most X servers, even those which do not have specific configuration
files, can use the directory specified by SERVER_MISC_CONFIG_PATH when
they have either the XSECURITY or XSELINUX extensions enabled, or when
support for DTRACE is enabled at build time, because this is also where
the "protocol.txt" file is searched for at runtime.
Unfortunately, the SERVER_MISC_CONFIG_PATH is set from serverconfigdir
which is hardcoded in the build system to "$prefix/$libdir/xorg", and
all X server builds share the same path.
That makes it harder for different X servers such as Xwayland to install
in the same path without sharing the same server configuration path
(and hence the same "protocol.txt" file).
Allow for the customization of server configuration path from the build
options so that different X servers can use completely different and
independent paths.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
(cherry picked from commit 411a61f571)
When running fullscreen, if an X11 client has changed the resolution,
Xwayland is using a viewport to emulate the expected resolution.
When changing focus, the Wayland compositor will send a configure event
with the actual surface size, not the size of the emulated XRandR
resolution.
As a result, changing focus while XRandR emulation (and hence the
viewport) is active in Xwayland will revert the resolution to the actual
output size, defeating the XRandR emulation.
To avoid that issue, only change the size when not running fullscreen.
Fixes: 53b6d4db7 - xwayland: Apply root toplevel configure dimensions
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Kenny Levinsen <kl@kl.wtf>
(cherry picked from commit a797776ff2)
Whenever the output configuration changes, if Xwayland is running
fullscreen, we may need to update the viewport in use or even update the
output on which Xwayland is currently running fullscreen.
Add a new helper function xwl_window_rootful_update_fullscreen() that
will recompute the fullscreen state and the viewport setup so that the
fullscreen Xwayland rootful window matches the new setup.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Kenny Levinsen <kl@kl.wtf>
(cherry picked from commit 73b9ff53c3)
The handling of appending/prepending properties was incorrect, with at
least two bugs: the property length was set to the length of the new
part only, i.e. appending or prepending N elements to a property with P
existing elements always resulted in the property having N elements
instead of N + P.
Second, when pre-pending a value to a property, the offset for the old
values was incorrect, leaving the new property with potentially
uninitalized values and/or resulting in OOB memory writes.
For example, prepending a 3 element value to a 5 element property would
result in this 8 value array:
[N, N, N, ?, ?, P, P, P ] P, P
^OOB write
The XI2 code is a copy/paste of the RandR code, so the bug exists in
both.
CVE-2023-5367, ZDI-CAN-22153
This vulnerability was discovered by:
Jan-Niklas Sohn working with Trend Micro Zero Day Initiative
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
(cherry picked from commit 541ab2ecd4)
This is more portable than libbsd as everything Just Works, even on BSD systems,
and is the recommended method of consuming libbsd nowadays.
It also helpfully lets things work with glibc-provided functions for new
enough glibc.
Closes: https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/973
Co-authored-by: Guillem Jover <guillem@hadrons.org>
Signed-off-by: Sam James <sam@gentoo.org>
(cherry picked from commit 94945a5274)
Xwayland maintains a connection to EI up for 10 minutes after an X11
client has vanished, to avoid going through the connection phase every
time a short lived X11 client comes and goes.
However, if the EI client gets freed (through some other event, e.g. the
user decides to terminate the EI session), Xwayland would still keep the
callback alive and end up trying to free an already freed EI client:
Invalid read of size 4
at 0x4C5E6F9: object_unref (util-object.h:89)
by 0x4C5E6F9: ei_unref (libei.c:77)
by 0x429525: free_ei (xwayland-xtest.c:224)
by 0x429A6E: disconnect_timer_cb (xwayland-xtest.c:404)
by 0x5E63FF: DoTimer (WaitFor.c:276)
by 0x5E6463: DoTimers (WaitFor.c:290)
by 0x5E6164: check_timers (WaitFor.c:133)
by 0x5E61E9: WaitForSomething (WaitFor.c:195)
by 0x4AD50E: Dispatch (dispatch.c:487)
by 0x4BBA0B: dix_main (main.c:272)
by 0x43615D: main (stubmain.c:34)
Address 0x15cc6ee8 is 8 bytes inside a block of size 240 free'd
at 0x48452AC: free (vg_replace_malloc.c:974)
by 0x4C5E729: object_destroy (util-object.h:73)
by 0x4C5E729: object_unref (util-object.h:91)
by 0x4C5E729: ei_unref (libei.c:77)
by 0x429525: free_ei (xwayland-xtest.c:224)
by 0x42A946: xwl_handle_ei_event (xwayland-xtest.c:804)
by 0x5EA977: HandleNotifyFd (connection.c:809)
by 0x5EE8E3: ospoll_wait (ospoll.c:657)
by 0x5E624D: WaitForSomething (WaitFor.c:208)
by 0x4AD50E: Dispatch (dispatch.c:487)
by 0x4BBA0B: dix_main (main.c:272)
by 0x43615D: main (stubmain.c:34)
Block was alloc'd at
at 0x484782C: calloc (vg_replace_malloc.c:1554)
by 0x4C5E777: ei_create (libei.c:73)
by 0x4C5E777: ei_create_context (libei.c:97)
by 0x42994B: setup_ei (xwayland-xtest.c:366)
by 0x42A383: xwayland_xtest_send_events (xwayland-xtest.c:658)
by 0x54ED4C: ProcXTestFakeInput (xtest.c:441)
by 0x54EE56: ProcXTestDispatch (xtest.c:475)
by 0x4AD6E6: Dispatch (dispatch.c:546)
by 0x4BBA0B: dix_main (main.c:272)
by 0x43615D: main (stubmain.c:34)
To avoid that issue, make sure to cancel the timer as soon as a EI
client is freed.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
See-also: https://bugzilla.redhat.com/2243076
(cherry picked from commit 9617de733b)
If we fail to setup EI, give up on using EI for XTEST and restore the
default XTEST handlers.
This happens when neither the portal nor the socket backends are usable.
This does not affect the portal operation though, if the user choose not
to allow a particular client, Xwayland would continue to use EI.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Joshua Ashton <joshua@froggi.es>
(cherry picked from commit 6b56ae68e5)
With EI support wired to XTEST, and oeffis being enabled unconditionally
means that Xwayland will always go through the XDG portal for XTEST when
supported.
While this the intended behavior for the general use case of Xwayland
running rootless on a desktop compositor, that breaks when Xwayland is
running on a nested compositor, because the portal is for the entire
session and not limited to the nested Wayland compositor.
Xwayland itself, as a regular Wayland client, has no way to tell that it
is running on a nested compositor.
So to keep backward compatibility with existing (and also common) use
cases such as nested compositors, best is to disable support for the XDG
portal by default, and add a new command line option "-enable-ei-portal"
for the Wayland compositors (who spawn Xwayland rootless) to explicitly
enable support for the input emulation XDG portal in Xwayland.
A Wayland compositor running nested should not use that command line
option with Xwayland.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Joshua Ashton <joshua@froggi.es>
Fixes: a1333342 - xwayland: Add XTEST support using EIS
Closes: https://gitlab.freedesktop.org/xorg/xserver/-/issues/1586
See-also: https://gitlab.gnome.org/GNOME/mutter/-/issues/3047
(cherry picked from commit cfcbb075c2)
Some drivers might not support explicit format modifiers. On these
drivers `gbm_bo_create_with_modifiers()` will fail and the
`gbm_bo_create()` code path will be used instead.
In this case, if the LINEAR modifier is advertised (and the INVALID
modifier is not) add the `GBM_BO_USE_LINEAR` flag.
Closes: https://gitlab.freedesktop.org/xorg/xserver/-/issues/1438
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Signed-off-by: José Expósito's avatarJosé Expósito <jexposit@redhat.com>
(cherry picked from commit 287638db59)
This can happen if the window has never completed a Present operation.
Fixes: 4230176080 ("xwayland/present: Embed present_vblank_rec in xwl_present_event")
(cherry picked from commit 32c5b2c044)
If there is no quads to draw, then we have a possibility to call
glDrawElements with type as zero, which will generate
GL_INVALID_ENUM error. While this error is harmless, it is annoying.
Signed-off-by: Konstantin <ria.freelander@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
(cherry picked from commit baaddf47d5)
If a client tries to send XTEST events while there is no sendEventsProc
defined for the given device, Xwayland would call into 0x0 and crash.
Make sure the handler is defined before trying to use it, to avoid the
crash.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
Closes: https://gitlab.freedesktop.org/xorg/xserver/-/issues/1574
(cherry picked from commit e820030de2)
We specify a sensible default geometry for decorated rootful windows,
but not for undecorated ones. Make the default geometry apply to rootful
windows in general.
Signed-off-by: Kenny Levinsen <kl@kl.wtf>
(cherry picked from commit 8128a21554)
While we now have support for resize of the root window through
libdecor, we still ignore toplevel configure dimensions when libdecor is
not in use. This ignores user intent in many Wayland servers, and some
xdg_toplevel states when active have strong requirements for adherence
to configure dimensions.
Resize in response to xdg_toplevel configure dimensions like we do for
libdecor configure events.
Signed-off-by: Kenny Levinsen <kl@kl.wtf>
(cherry picked from commit 53b6d4db7e)
The upcoming handling of plain xdg_toplevel.configure events will need
to use the xwl_window resize helper. Move it outside XWL_HAS_LIBDECOR,
move the remaining dimension logic from handle_libdecor_configure into
it and update the name accordingly.
Signed-off-by: Kenny Levinsen <kl@kl.wtf>
(cherry picked from commit 4f869c6eda)
