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
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>
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>
The Wayland interfaces have a "name" field that we can use instead of
hardcoding their name.
Change the code to use that name instead of the static strings.
This was inspired by a similar change in mutter by Robert Mader
<robert.mader@collabora.com>.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Peter Hutterer <peter.hutterer@who-t.net>
Xwayland offers a way for the window and compositing manager to hold the
surface commits through an X11 property _XWAYLAND_ALLOW_COMMITS.
Xwayland, however, does not actually check if the X11 client changing
the value of that property is indeed the X11 window manager, so any X11
client can potentially interfere with the Wayland surface mechanism.
Restrict access to the _XWAYLAND_ALLOW_COMMITS property to read-only,
except for the X11 window manager and the Xserver itself.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Pekka Paalanen <pekka.paalanen@collabora.com>
This is preparation work to restrict access to Xwayland properties.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Pekka Paalanen <pekka.paalanen@collabora.com>
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>
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>
This code is almost entirely ddx-agnostic already, and I'd like to use
it from the other EGL glamor consumers. Which, right now that's just
Xorg, but soon it'll be Xephyr too.
This commit adds an ability to store a glvnd vendor in Glamor
structures, which can be used for initialize some vendor-based values
without hooking into DDX internals. Also this adds setting this value
into Xorg and Xwayland
Signed-off-by: Konstantin Pugin <ria.freelander@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Acked-by: Emma Anholt <emma@anholt.net>
This replaces int glamor parameter with a new enum to be more clean
and prepare for more glamor options.
No functional change.
Signed-off-by: Konstantin <ria.freelander@gmail.com>
On Mesa, when we request GBM_FORMAT_XRGB8888, driver set surface storage
format to GL_RGB8, which breaks GL ES rendering (on any GL ES version).
If we force set gbm_format to GBM_FORMAT_ARGB8888, then rendering
will happen and working.
Fixes#1288Fixes#1356
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
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>
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
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>
This can happen if the window has never completed a Present operation.
Fixes: 4230176080 ("xwayland/present: Embed present_vblank_rec in xwl_present_event")
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>
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>
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>
When handling libdecor configure, we first update our xwl output and
screen if dimensions differ from the current xwl_screen, and then commit
a new libdecor frame which acknowledges the xdg_surface.configure event.
If the initial configure events contains non-zero dimensions, we will
update the xwl output before acknowledging the initial configure. As we
attach a buffer and commit the surface when updating the output, this
leads to a protocol error.
Instead, move the surface commit till the end of the configure handler
so it always happens after the ack.
Signed-off-by: Kenny Levinsen <kl@kl.wtf>
Similar to commit 94deed272 - " xwayland: Use sensible defaults for
rootful size", mark fullscreen mode as fixed so that the actual monitor
layout is not reflected in the single fullscreen rootful window.
Without this, if "-fullscreen" is used without "-geometry", the XRandR
configuration is taken from the compositor via wl_output/xdg-output and
cannot be changed by the X11 clients.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Enforce sensible min/max values for the window size when using libdecor.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
This is to avoid repeating the same code in two places.
This is essentially a cosmetic change, not a functional change.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Allow passing an optional libdecor configuration pointer to
xwl_window_update_libdecor_size() so that we can reuse it from more than
one place and avoid duplicating that code.
No functional change.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
The configure handler in libdecor is triggered any time a new
configuration is received.
According to the documentation from libdecor, an application should
respond to that event by creating a suitable libdecor_state, and apply
it using libdecor_frame_commit().
So we ought to attach a new buffer matching the new size and commit
the Wayland surface.
The actual content of the window does not need to be explicitly
repainted, that occurs through the call to SetRootClip():
xwl_output_set_mode_fixed()
-> update_screen_size()
-> SetRootClip()
-> miHandleValidateExposures()
-> miWindowExposures()
-> miPaintWindow()
This fixes an issue with mutter where maximizing a window and then
switching to another window would sometimes resize the Xwayland window
back to its pre-maximized size, or with Weston where the Xwayland window
would initially show up black until the pointer moves to the window.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
This moves the code which updates the XRandR modes and sets the root
window size to its own function.
This preparation work for the next commit, no functional change.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
The configure handler for libdecor, namely handle_libdecor_configure(),
is where both the content and the decorations get resized (when needed).
If for any reason, the actual size of the Xwayland screen fails to be
updated, we would still appy the expected size rather than the actual
one for the libdecor state.
To avoid this, use the actual xwl_screen width/height for the libdecor
state.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
For libdecor, we will have to attach a new buffer and commit from two
different handlers (libdecor configure and commit).
Having xwl_window_attach_buffer() separate from xwl_window_post_damage()
is to allow for that.
This commit should not introduce any functional change.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
glamor ensures that a depth 32 pixmap backing a depth 24 window contains
fully opaque alpha channel values for the window's pixels, so we can
allow this without implicit redirection, saving pixmap storage and
intermediate copies.
Second attempt, after fixing a few regressions from the first attempt.
If "-decorate" is used but no "-geometry" is specified, Xwayland rootful
would take its size from the actual Wayland outputs combined.
That is not practical, especially when using multiple outputs, as the
resulting Xwayland window would be much larger than a single monitor.
To avoid that, set a sensible default size for the Xwayland decorate
window, using 640x480 to match what Xephyr does.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
libdecor support seems quite stable, no need to mark that experimental.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
By default, the Xwayland window in rootful mode was not resizable.
Make the Xwayland window resizable using libdecor in rootful mode.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
This is preparation work for making Xwayland rootful resizeable.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
glamor ensures that a depth 32 pixmap backing a depth 24 window contains
fully opaque alpha channel values for the window's pixels, so we can
allow this without implicit redirection, saving pixmap storage and
intermediate copies.
This will be needed with the next commit: If a child window completely
obscures a toplevel ancestor of different depth, the child window can
use page flipping only if the depth of the presented pixmap matches that
of the window's backing pixmap, or the former may contain pixel values
which are not suitable for the toplevel window's depth.
With optional EI support in Xwayland, we would route XTEST events to EI
so that they get actually emulated in the Wayland compositor.
However, this implies that EI is actually supported in various places,
including the Wayland compositor of course. If, for whatever reason, we
fail to use EI, the actual XTEST events will be dropped.
That might be seen as a regression, as previously those would go through
the usual X11 processing of events and might have worked with X11 native
clients.
So, to keep backward compatibility, fallback to the plain old XTEST
method if EI is not available or not usable.
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
This adds support for XTEST in Xwayland using EIS, the emulated input
library [1].
To differentiate between X11 clients using XTEST, initiate a EI context
for each client and use the actual client name, from its command
line.
When an X11 client first tries to use XTEST to generate emulated input
events, a new connection to libEI is initiated by Xwayland on behalf
of the X11 client.
During that connection phase, the EI server will not be accepting
events until the emulated device is actually created, meaning that any
XTEST request from the X11 client will be discarded until the EI server
is willing to accept events.
To avoid that issue, add an event queue in Xwayland per X11 client that
will keep those requests, and dequeue them as soon as the EI server is
ready, i.e. once the EI device is added.
If the X11 client disconnects from the Xserver before the EI server is
ready, or if the connection is closed by the EI server, those events are
discarded and the queue cleared from any pending events.
For 10 minutes after the client disconnects, keep the internal struct
alive. If a client with the same commandline arguments connects again,
re-use the same struct. This means we are faster with the events the
second time around but it also allows the EIS server to pause individual
clients that keep sending intermittent events and disconnect immediately
(e.g. it'd be possible to pause xdtotool while an authentication prompt
is active).
[1] https://gitlab.freedesktop.org/libinput/libei
Thanks to Jan Beich <jbeich@FreeBSD.org> for fixing the build on BSD.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Co-authored-by: Olivier Fourdan <ofourdan@redhat.com>
Co-authored-by: David Redondo <kde@david-redondo.de>
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Nothing should be relying on this anymore, so use a counter like other
places in the tree instead. This ensures that the event_id doesn't get
cast back into a pointer again in future, and also may be slightly less
confusing in cases where calloc reuses an address as debug logs would
show the same event_id for those but now they will be distinct.
Signed-off-by: Jessica Clarke <jrtc27@jrtc27.com>
On traditional 32-bit and 64-bit architectures, uint64_t can be abused
to hold a uintptr_t and be cast back to a valid pointer. However, on
CHERI, and thus Arm's Morello prototype, pointers are capabilities,
which contain a traditional address alongside additional metadata,
including a tag bit that ensures it cannot be forged (the only way to
get a capability with the tag bit set is by using instructions that take
in another valid capability with sufficient bounds/permissions/etc for
the request, and any other operation, like overwriting individual bytes
in memory, will give a capability whose tag is clear). Casting a pointer
to a uintptr_t is fine as uintptr_t is represented as a capability, but
casting to a uint64_t yields just the address, losing the metadata and
tag. Thus, when cast back to a uintptr_t, the capability remains invalid
and faults on any attempt to dereference.
As with various other places in the tree, address this by searching for
the pointer in a list so that we no longer rely on this undefined
behaviour.
Signed-off-by: Jessica Clarke <jrtc27@jrtc27.com>
All these arguments other than damage come from the vblank itself so
passing the vblank simplifies the caller. Moreover, we pass the event_id
solely so we can get back to the event, which is just the (extended)
vblank, so passing the vblank avoids that round trip.
Signed-off-by: Jessica Clarke <jrtc27@jrtc27.com>
Warren Togami
Olivier Fourdan
Adam Jackson
Konstantin Pugin
José Expósito
Michel Dänzer
Kenny Levinsen
Russell Chou
Peter Hutterer
Xaver Hugl
Jessica Clarke