The cleanup function for GBM is called on the various error paths.
Once xwl_glamor_gbm_cleanup() has been called, GBM support is no longer
usable (and the corresponding data structures are freed), so there is
no way we can keep using GLAMOR after that point.
Make sure to explicitly disable GLAMOR support in that case, so we do
not crash later on trying to use GBM.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1768>
Consider the following keymap:
```xkb
xkb_keymap {
xkb_keycodes {
<compose> = 135;
};
xkb_symbols {
key <compose> {
[ SetGroup(group = +1) ]
};
};
};
```
When the user presses the compose key, the following happens:
1. The compositor forwards the key to Xwayland.
2. Xwayland executes the SetGroup action and sets the base_group to 1
and the effective group to 1.
3. The compositor updates its own state and sends the effective group,
1, to Xwayland.
4. Xwayland sets the locked group to 1 and the effective group to
1 + 1 = 2.
This is wrong since pressing compose should set the effective group to 1
but to X applications the effective group appears to be 2.
This commit makes it so that Xwayland completely ignores the key
behaviors and actions of the keymap and only updates the modifier and
group components in response to the wayland modifiers events.
Signed-off-by: Julian Orth <ju.orth@gmail.com>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1758>
Currently, Xwayland creates a pixmap backed by shared memory buffer as
soon as an X11 cursor is realized, which is destroyed when the cursor is
eventually unrealized.
If an X11 client is leaking cursors, Xwayland will be creating new
pixmaps continuously, which will eventually cause an error once the
limit is reached, and get Xwayland killed.
However, we do not need the shared memory buffer to stay around, we
already have the buffer retention mechanism which will take care of
keeping the buffer around until the Wayland compositor is done with it,
so we could just create and destroy the pixmap as needed when setting
the cursor.
That would not fix the leak in the X11 application, yet that would
mitigate the risk of Xwayland being killed by reaching the shared memory
limits, until the client itself reaches the limit of X11 resources.
v2: Don't increase the pixmap refcnt to destroy it just after (Michel)
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Suggested-by: Michel Dänzer <michel@daenzer.net>
See-also: https://gitlab.freedesktop.org/xorg/xserver/-/issues/1773
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1754>
For the Wayland compositors who do not implement XkbBellNotifyMask but
support the Wayland protocol xdg-system-bell, use that to ring the
bell.
v2: Be more selective on the device, make sure it's a keyboard and it
has core events.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1742>
eglCreateSyncKHR takes ownership of the file descriptor. Noticed by
inspection.
While we're at it, move the fence_fd declaration to the scope where
it's used.
Last but not least, close the fd in xwl_glamor_wait_fence when bailing
before calling eglCreateSyncKHR, and document that it takes ownership.
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1712>
The symbol controls whether to include dix-config.h, and it's always set,
thus we don't need it (and dozens of ifdef's) anymore.
This commit only removes them from our own source files, where we can
guarantee that dix-config.h is present - leaving the (potentially exported)
headers untouched.
Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>
Even if the manual redirect didn't result in the surface window we
wanted, it might be different again without the redirect.
I don't know of any specific scenario hitting this though, it's more of
a theoretical safeguard.
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1637>
using EGL (e.g., eglQueryString, epoxy_has_egl_extension)
before establishing this connection
enables the GBM/EGL implementation to potentially consume the
WAYLAND_SOCKET fd, which, if closed, will cause the compositor
to kill this xserver
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1646>
Most of the support for such OS'es was removed in 2010 for
xorg-server-1.10.0, but a few bits lingered on, and a few
comments were left out-of-date.
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1667>
The vidmode extension emulation in Xwayland reports the modeline based
on the current mode.
To do so, it searches for the mode using `xwl_output_find_mode(-1, -1)`
which is supposed to return the current mode, whatever that mode is.
With XRandR emulation, in rootless mode, the default value is the mode
at index 0. That assumption, however is not true when running rootful.
That means that the vidmode extension will always return the highest
mode available, which is 5120x2880, with Xwayland running rootful:
$ xwayland-run -geometry 1024x768 -- xvidtune -show
"5120x2880" 1276.50 5120 5560 6128 7136 2880 2883 2888 2982 -hsync +vsync
Luckily, when Xwayland is running rootful, we have the current mode size
conveniently stored in dedicated fields of the xwayland output struct,
so we can use that to search for the right mode being used and report
that through the vidmode extension:
$ xwayland-run -geometry 1024x768 -- xvidtune -show
"1024x768" 63.50 1024 1064 1176 1328 768 771 775 798 -hsync +vsync
That fixes legacy games using the vidmode extension and rendering at the
wrong size when running within Xwayland rootful.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1641>
The code here assumed a `leave` event always occurs between two `enter`
events. On Sway (and presumably other compositors) this happens even if
the client has destroyed the `wl_surface`, but the client gets a null
`surface` here. (Which presumably on on the wire is the id of the
destroyed surface.)
This seems like a bad thing to rely on, and is easy to avoid. But if
this is correct to assume, the Wayland protocol should be explicit about
this.
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1527>
Now that we won't enable DRI3 if <sys/eventfd.h> is not available, there
is not point in trying to include that header without DRI3.
That allows to build Xwayland with GLAMOR enabled (without DRI3) on
platforms which do not support eventfd.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1523>
Currently, we would start dequeuing events as soon as a device is
resumed, regardless of its capabilities.
If the capabilities are not available, we would just fallback to the
regular XTEST code path and not use input emulation.
As a result, it is very likely that we shall lose the first events until
the compositor resumes first a device with the requested capabilities.
To avoid that issue, start emulating only once we have the requested
capabilities, if they match the seat capabilities.
Closes: https://gitlab.freedesktop.org/xorg/xserver/-/issues/1732
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1631>
This is a small code refactoring to help with clarity, simply move the
code from the switch case for device resume to a dedicated function.
No functional change.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1631>
This has been nothing but an alias for two decades now (somewhere in R6.6),
so there doesn't seem to be any practical need for this indirection.
The macro still needs to remain, as long as (external) drivers still using it.
Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1529>
This has been nothing but an alias for two decades now (somewhere in R6.6),
so there doesn't seem to be any practical need for this indirection.
The macro still needs to remain, as long as (external) drivers still using it.
Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1529>
This has been nothing but an alias for two decades now (somewhere in R6.6),
so there doesn't seem to be any practical need for this indirection.
The macro still needs to remain, as long as (external) drivers still using it.
Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1529>
The xnfreallocarray was added along (and just as an alias to) XNFreallocarray
back a decade ago. It's just used in a few places and it's only saves us from
passing the first parameter (NULL), so the actual benefit isn't really huge.
No (known) driver is using it, so the macro can be dropped entirely.
Fixes: ae75d50395
Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1529>
This has been nothing but an alias for two decades now (somewhere in R6.6),
so there doesn't seem to be any practical need for this indirection.
The macro still needs to remain, as long as (external) drivers still using it.
Fixes: ded6147bfb
Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1529>
Currently, we would log only the event type, use the libei API to also
log the name in plain text, so we can quickly identify the events we're
missing out.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1620>
Since commit d370f1e58, Xwayland can optionally be started rootful and
fullscreen.
To do so, it will setup a viewport to scale the root window to match the
size of the output.
However, if the rootful Xwayland window receives an xdg-surface configure
event before the output definition is complete, as with e.g. the labwc
Wayland compositor, we might end up trying to setup a viewport with a
destination size of 0x0 which is a protocol violation, and that kills
Xwayland.
To avoid that issue, only setup the viewport if the output size is
meaningful.
Also, please note that once the output definition is complete, i.e. when
the "done" event is eventually received, we shall recompute the size for
fullscreen again, hence achieving the desired fullscreen state.
Fixes: d370f1e58 - xwayland: add fullscreen mode for rootful
Closes: https://gitlab.freedesktop.org/xorg/xserver/-/issues/1717
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1621>
Several places using _X_ATTRIBUTE_PRINTF macro from X11/Xfuncproto.h
but missing to include it, so it depends on other headers whether it's
included by mere accident, which quickly causes trouble if include order
changes. Cleaning that up by adding explicit include statements.
Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1580>
Clears -Wcalloc-transposed-args warnings from gcc 14.1, such as:
../dix/main.c:165:42: warning: ‘calloc’ sizes specified with ‘sizeof’ in the
earlier argument and not in the later argument [-Wcalloc-transposed-args]
165 | serverClient = calloc(sizeof(ClientRec), 1);
| ^~~~~~~~~
../dix/main.c:165:42: note: earlier argument should specify number of
elements, later size of each element
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1606>
Since commit 792758fa ("xwayland: Update lost focus on deactivation"),
in rootful mode, if we receive an "activated" state from xdg-shell
indicating that the surface is no longer active, we shall end up calling
xwl_seat_leave_ptr().
But xwl_seat_leave_ptr() does not actually check whether the seat has
pointer capabilities, and if not, get_pointer_device() will return NULL.
As a result, we would crash using a NULL pointer in GetMaster(). This
typically can happen when using Xwayland rootful on headless compositors
such as "cage" which do not advertise any capabilities for the seat.
To avoid the issue, simply check whether get_pointer_device() returns a
valid non-null device and bail out early otherwise.
Fixes: 792758fa - xwayland: Update lost focus on deactivation
Closes: https://gitlab.freedesktop.org/xorg/xserver/-/issues/1700
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1590>
With explicit buffer synchronization in use, the window buffers use a
file descriptor for event notification to keep the buffer alive for
synchronization purpose.
When running rootful, the root window (which is visible) is destroyed
directly from the resource manager on server reset, and the window
buffer's eventfd will trigger after the window is destroyed, leading to
a use after free and a crash of the xserver.
To avoid the issue, check whether the window being destroyed is the root
window in rootless mode, and make sure to force the disposal of the
window buffers in that case.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Closes: https://gitlab.freedesktop.org/xorg/xserver/-/issues/1699
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1589>
For cases (to come) where we would want to force the disposal of the
window buffers, add a parameter to force the disposal by calling
dispose() directly instead of maybe_dispose().
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1589>
Right now, we would dispose the xwl_window and all the data associated
with it on unrealize.
But not all window destruction go through the unrealize code path, for
example when the root window (running rootful) is destroyed from the
resource manager on server reset, we do not get to the unrealize window
step, but straight to destroy window.
As a result, we are leaking the xwl_window and all the data associated
with it, for example:
| 65,536 bytes in 1 blocks are possibly lost in loss record 12,462 of 12,488
| at 0x484A0FC: calloc (vg_replace_malloc.c:1675)
| by 0x48B661C: UnknownInlinedFun (pixman-bits-image.c:1273)
| by 0x48B661C: _pixman_bits_image_init (pixman-bits-image.c:1296)
| by 0x48B6754: create_bits_image_internal (pixman-bits-image.c:1349)
| by 0x64180DE: UnknownInlinedFun (cairo-image-surface.c:380)
| by 0x64180DE: UnknownInlinedFun (cairo-image-surface.c:366)
| by 0x64180DE: cairo_image_surface_create (cairo-image-surface.c:432)
| by 0x6346B44: UnknownInlinedFun (libdecor-gtk.c:467)
| by 0x6346B44: libdecor_plugin_gtk_frame_new (libdecor-gtk.c:546)
| by 0x4B7F297: libdecor_decorate (libdecor.c:559)
| by 0x42C6F3: xwl_create_root_surface (xwayland-window.c:1266)
| by 0x42CD97: ensure_surface_for_window (xwayland-window.c:1466)
| by 0x42D0D1: xwl_realize_window (xwayland-window.c:1560)
| by 0x50858F: compRealizeWindow (compwindow.c:279)
| by 0x4FF2A2: MapWindow (window.c:2706)
| by 0x4F9E7F: InitRootWindow (window.c:697)
To avoid that issue, check whether there is still an xwl_window
associated with the X11 window on destroy, and if that's the case,
dispose the xwl_window.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1589>
The wp_linux_drm_syncobj_v1 protocol states that :
| If at surface commit time there is a pending buffer attached but no
| pending release timeline point set, the no_release_point protocol
| error is raised.
So we need to set a release timeline point in any case from the swap
pixmap routine, even for the early out code paths.
Failing to do so may cause a Wayland protocol error that is fatal to the
Wayland client, in this case Xwayland:
| wp_linux_drm_syncobj_surface_v1: error 4: No Acquire point provided
| (EE) failed to dispatch Wayland events: Protocol error
Closes: https://gitlab.freedesktop.org/xorg/xserver/-/issues/1688
Fixes: 87bf2cafcc - xwayland: add support for wp_linux_drm_syncobj_v1
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1571>
Move the code which takes care of submitting pixmaps and the
synchronization points to its own function.
This will allow to reuse that code from different code path.
No functional change intended.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1571>
The function xwl_window_swap_pixmap() can be called from two places,
either from xwl_window_attach_buffer() or from damage_report().
When called from xwl_window_attach_buffer(), the new buffer is attached
and the surface committed.
However, when called from damage_report(), a new buffer might not be
attached before the surface is committed.
That's fine with implicit synchronization, but if we use explicit
synchronization, committing a surface without a new buffer attached but
with a release timeline point set is a protocol error:
| If at surface commit time there is a pending release timeline point
| set but no pending buffer attached, a no_buffer error is raised.
To avoid such an issue, add a new parameter to xwl_window_swap_pixmap()
to hint whether it should set the synchronization points, and have the
synchronization points set only from xwl_window_attach_buffer().
v2: Rename param to handle_sync (Michel)
Suggested-by: Michel Dänzer <mdaenzer@redhat.com>
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Part-of: <https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/1571>
Enrico Weigelt, metux IT consult
Olivier Fourdan
Michel Dänzer
Julian Orth
Michel Dänzer
YaoBing Xiao
Mike Blumenkrantz
Alan Coopersmith
Ian Douglas Scott