Followup on 4e4898e90f.
If a view quickly maps and unmaps repeatedly, there will be multiple
destroyed containers with same view in a single transaction. Each of
these containers will then try to destroy this view, resulting in use
after free.
The container should only destroy the view if the view still belongs
to the container.
Simple reproducer: couple XMapWindow + XUnmapWindow in a loop followed
by XDestroyWindow.
See #6605
(cherry picked from commit f92329701b0983ec41fec29d3abc5c751cbe4a28)
We currently track the focus of a seat in two ways: we use a list called
focus_stack to track the order in which nodes have been focused, with
the first node representing what's currently focused, and we use a
variable called has_focus to indicate whether anything has focus--i.e.
whether we should actually treat that first node as focused at any given
time.
In a number of places, we treat has_focus as implying that a focused
node exists. If it's true, we attempt to dereference the return value of
seat_get_focus(), our helper function for getting the first node in
focus_list, with no further checks. But this isn't quite correct with
the current implementation of seat_get_focus(): not only does it return
NULL when has_focus is false, it also returns NULL when focus_stack
contains no items.
In most cases, focus_stack never becomes empty and so this doesn't
matter at all. Since focus_stack stores a history of focused nodes, we
rarely remove nodes from it. The exception to this is when a node itself
goes away. In that case, we call seat_node_destroy() to remove it from
focus_stack and free it. But we don't unset has_focus if we've removed
the final node! This lets us get into a state where has_focus is true
but seat_get_focus() returns NULL, leading to a segfault when we try to
dereference it.
Fix the issue both by updating has_focus in seat_node_destroy() and by
adding an assertion in seat_get_focus() that ensures focus_stack and
has_focus are in sync, which will make it easier to track down similar
issues in the future.
Fixes#6395.
[1] There's some discussion in #1585 from when this was implemented
about whether has_focus is actually necessary; it's possible we could
remove it entirely, but for the moment this is the architecture we have.
(cherry picked from commit 921b0a863382b70234aeb4bd589c10328e9ff042)
cairo_image_surface_create can fail, e.g. when running out of
memory or when the size is too big. Avoid crashing in this case.
Closes: https://github.com/swaywm/sway/issues/6531
(cherry picked from commit 59aebaa5f9f3afe9cdfbb0d37c4dc631690da3b9)
Now output_begin_destroy emits the node::destroy event similar to
workspace_begin_destroy. It currently has no listeners, since they
listen to output::disable or wlr_output::destroy instead.
We use the headless backend to create a special fallback output
used when no other output is connected. However this messes up the
"real" headless output names users have come to expect (e.g.
currently the first headless output will be named "HEADLESS-2"
instead of "HEADLESS-1").
Fix this by setting the output name with [1].
[1]: https://gitlab.freedesktop.org/wlroots/wlroots/-/merge_requests/3395
sway-bar(5) says:
> For compatibility with i3, bar mode <mode> [<bar-id>] syntax is
> supported along with the sway only bar <bar-id> mode <mode> syntax.
while the actual behavior is that `bar_cmd_mode` ignores already
selected `config->current_bar` and applies the change to all the
configured bars.
This makes it possible to hint to the renderer and backends how many
bits per channel the buffers that the compositor draws windows onto
should have. Renderers and backends may deviate from this if they
do not support the formats with higher bit depth.
Proprietary drivers require --unsupported-gpu to be allowed, and IPCs
require no option to be passed.
The only way to satisfy both is to run IPCs before checking for
proprietary drivers.
Wayland compositors handle many file descriptors: client
connections, DMA-BUFs, sync_files, wl_data_device pipes, and so
on. Bump the limit to the max.
Closes: https://github.com/swaywm/sway/issues/6285
Add a subcommand for `smart_gaps` that enables outer gaps only
on workspaces with exactly one visible child.
Also add documentation for `smart_gaps toggle`.
previously, fullscreen global containers would grab cursor input
even if a shell-layer surface was on top of it
related issue: https://github.com/swaywm/sway/issues/6501
If the focused container is floating by itself, create a new container
in tiling mode as a sibling of the inactive focused container instead of
creating it as a sibling of everything that is in tiling mode in that
workspace. This is the i3 behavior.
seat_get_focus_inactive_floating and seat_get_focus_inactive_tiling do
not always return a view, so get the previously focused view from the
container with seat_get_focus_inactive_view. This is the i3 behavior.
If the destroyed xwayland view is in transaction, it won't
be destroyed immediately. wlr_xwayland_surface then becomes
dangling pointer.
Closes#6605Closes#5884
Nvidia has historically been a bad actor in the open-source graphics
ecosystem because they required a special EGLStreams code-path
instead of exposing the de-facto standard GBM API. However, with
their upcoming release they now support GBM as well.
This is a push in the right direction for Nvidia, so there's no
reason we should be more hostile to them than to any other proprietary
driver. Let's remove the --my-next-gpu-wont-be-nvidia flag, and advise
users to use --unsupported-gpu now.
Note, proprietary Nvidia drivers are still unsupported by the Sway
project (just like all other proprietary drivers).
Adds detection code to handle pci-*-platform-* strings
in ID_PATH
References: https://github.com/swaywm/sway/issues/6590
Signed-off-by: Jari Ronkainen <ronchaine@gmail.com>
Commit 152a559e replaced the view pointer in the inhibitor struct with a
pointer to the wlr_inhibitor for application inhibitors. But this was not
changed in the sway_idle_inhibit_v1_application_inhibitor_for_view function.
This caused a bug in the sway tree view where the application inhibitor is
always "none".
Clang 13 reports:
../sway/commands.c:470:23: error: variable 'context' set but not used
[-Werror,-Wunused-but-set-variable]
enum command_context context = 0;
^
Last use of was removed in commit 1d3681f521.
Downstream PR: https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=258813
This commit makes sure the extents are kept up-to-date, fixes not
damaging the surface if its layer shell-specific state didn't change,
and adds a check if the layer shell-specific state didn't change but the
surface got mapped/unmapped, which could affect keyboard focus.
Prior to 62d90a8e, titlebar's font height (and other related values)
would change any time any titlebar's content changed, so these values
were recalculated each time any titlebar's content changed (or a new
titlebar was created).
However, since the above was merge, these values no longer change so
often and we only need to recalculate them when the configured font
changes (and stop calling `config_update_font_height` each time
titlebars are rendered).
This commit removes all the unecessary calls to this function and avoids
all those unecessary calculations. Whenever the font strays from the
default value, the `font` command is called, and it calls
`config_update_font_height`, which is enough to keep the value always up
to date.
I've also added a default value to the `font_baseline` config, since
otherwise that's zero for setups that don't explicitly specify a font.
This prevents sway from extending the desktop to i.e. VR headsets, and makes
them available for DRM leasing.
Non-desktop wlr_outputs will be offered through the wlr_drm_lease_v1_manager
interface for client to lease.
If the surface the pointer started to interact with is destroyed we also
want the seatop_down to end. In case a drag is initiated we receive a
call to handle_end.
This solves an issue where layer-shell items would not receive a button
release event when the pointer left them while being pressed. The
default seatop changes focus immediately while seatop_down defers any
focus changes until the pointer is released or seatop_down is destroyed.
The title itself and marks were being rendered by two very-similar yet
different functions, and any changes made to one had to be reflected on
the other.
This mostly prevents such oversights from happening, and keeps makes
sure we keep both consistent.
Use fixed titlebar heights. The default height is calculated based on
font metrics for the configured font and current locale.
Some testing with titles with emoji and CJK characters (which are
substantially higher in my setup) shows that the titlebars retain their
initial value, text does shift up or down, and all titlebars always
remain aligned.
Also drop some also now-unecessary title_height calculations.
Makes also needed to be updated, since they should be positioned with
the same rules.
Sometimes the preferred mode is not available due to hardware
constraints (e.g. GPU or cable bandwidth limitations). In these
cases it's better to fallback to lower modes than to end up with
a black screen.
When a layer surface shrinks we need to damage the area it previously
occupied, but we don't know the location of all its subsurfaces in the
previous state, so instead damage a rectangle that encloses the entire
previous extent.
The xdg-decoration protocol allows clients to request whether they want
to use server side decorations or client side decorations. Currently,
sway ignores this and always enforces whatever the server is currently
set to. Although tiled clients cannot be allowed to set borders, there
is no harm in listening requests from floating clients. Sidenote: also
fix an unrelated style error.
Losing the precision resulted in wlr_cursor and wlr_seat::pointer_state
getting out of sync during pointer motion in seatop_down.
Since the difference was always under 1 px, it was practically
impossible to notice in normal use.
But because of being out of sync, cursor_rebase would always end up
incorrectly calling wlr_seat_pointer_notify_motion from
seatop_default_begin (on releasing mouse button) which broke cursor
locking.
See #5405Closes#4632
When emulating touch, the simulating_pointer_from_touch field is
set to true. It's switched back to false when a touch_up event is
received. However we need to ensure we always send a wl_pointer.frame
event following a group of other wl_pointer events.
Since a touch_frame event is always guaranteed to come after a group
of touch events, unset simulating_pointer_from_touch in the touch_frame
handler instead of the touch_up handler. Add a new field to know whether
the touch_frame handler should stop emulation.
If HOME environment variable is not set, sway fails startup with a
segmentation fault due to null pointer dereference.
Also check calloc return value and only perform the fallback code when
really needed.
Signed-off-by: Tobias Stoeckmann <tobias@stoeckmann.org>
Add a subcommand for `gaps` that allows to toggle gaps at
runtime. This functionality is part of i3-gaps since [1]
but is missing in sway.
[1] https://github.com/Airblader/i3/pull/264
When setting the geometry from content for floating windows, the
coordinates for borders are normally taken into account. However in the
case of a floating fullscreen window, we should not be doing this. Since
the content of the container takes the space of the entire output, this
causes the calculated borders to neccesarily be outside of the output.
This later causes a problem when sending surface entrance events since
in a multi-monitor setup, the border coordinates will overlap with
another output despite the surface not actually being on that output at
all. The fix is to just ignore border coordinates for a floating
fullscreen container since fullscreen, of course, does not actually have
any borders. Fixes#6080.
get_current_time_msec is only used in cursor.c, so we can move it in and
make it static. This is primarily intended to avoid a symbol collision
with wlroots, which we unfortunately do not have a good solution for
yet.
This fixes the following scenario:
- Place a floating window so its border is right at the edge of the
screen
- Create a new split
- The border disappears
- Moving the window does not restore the border
Instead of disabling it for some workspace subcommands, this explicitly
calls it only in the 2 places it's actually needed: for switching to a
named or numbered workspace.
This extracts the code to a separate workspace_auto_back_and_forth
function.
It also removes the bool argument by adding an extra if statement at the call
site, and repurposes the no_auto_back_and_forth variable to
auto_back_and_forth for simpler understanding.
This forces no_auto_back_and_forth to true for `workspace
next_on_output` and `workspace prev_on_output` to keep parity with i3.
In i3, running next_on_output never changes focus to another output.
In Sway currently, with workspace_auto_back_and_forth set to yet,
running next_on_output on an output with only a single active workspace
will typically end up focussing the other output:
1. next_on_output focusses the current workspace, because it's the only
one
2. auto_back_and_forth focusses the last focussed workspace, because the
current workspace to focus is the current one. This will usually be on
the other monitor if the workspace there was last focussed.
Sway ignores SIGPIPE (by installing a SIG_IGN handler), in order to
“prevent IPC from crashing Sway”.
SIG_IGN handlers are the *only* signal handlers inherited in
sub-processes. As such, we should be a good citizen and restore the
SIGPIPE handler to its default handler.
Original bug report:
https://www.mail-archive.com/debian-bugs-dist@lists.debian.org/msg1806907.html
Sway ignores SIGPIPE (by installing a SIG_IGN handler), in order to
“prevent IPC from crashing Sway”.
SIG_IGN handlers are the *only* signal handlers inherited in
sub-processes. As such, we should be a good citizen and restore
the SIGPIPE handler to its default handler.
Original bug report:
https://www.mail-archive.com/debian-bugs-dist@lists.debian.org/msg1806907.html
Until now, swaybar did not have pango markup enabled by default, even if
the sway config had it on. This patch aims to mimic the i3 behavior, but
maintaining the functionality of the "pango_markup" sway config command.
Deferred commands are only run once, during sway startup. This means
that deferring seat attachment based on whether we are reading the
config prevents devices from being reattached to the correct seat during
a config reload. Instead, only defer if the config is not yet active.
Fixes#6048.
Implements functionality described in [1]. Please see the issue for a
video with a demonstration of the new behavior.
An issue is that titlebars cover up a significant portion of the top
edge drop area. The solution is simply to change the edge drop area
hitbox to start at the contents instead of the container.
[1] https://github.com/swaywm/sway/issues/6218
There was some unused code-paths for rendering surfaces with an
arbitrary rotation applied. This was imported from rootston.
Since we don't have plans to make use of this, remove it.
render_surface_iterator previously deduced the clip box from an optional
container passed with render data. This causes problems when offsets in
view geometry need to be compensated for in the clip dimensions.
Instead, prepare the clip box in render_view_toplevels where the offsets
are being applied, and compensate for them immediately.
A similar compensation is applied to render_saved_view.
Closes: https://github.com/swaywm/sway/issues/6223
These coordinates contain the all-time accumulated buffer attach point,
which is a way to perform incremental client-side initiated movement of
windows, intended as a way to maintain logical window positioning while
compensating for layout changes such as folding in a left side panel.
This value is not useful for implementing this feature, and break things
if they ever become non-zero. Their inclusion in calculations also tend
to cause confusion.
Remove usage of these coordinates, removing the ability for clients to
move themselves. This may again be supported if a better API is made
available from wlroots.
remove view from its own unmap event listener so when subsurfaces
link try to remove themselves they won't run into it.
This fixes the following ASAN use-after-free error on a build slightly
modified to instrument wl_list operations:
==71705==ERROR: AddressSanitizer: heap-use-after-free on address 0x6160000829a0 at pc 0x000000508eb7 bp 0x7ffec8fd8030 sp 0x7ffec8fd8028
WRITE of size 8 at 0x6160000829a0 thread T0
#0 0x508eb6 in wl_list_remove ../common/list.c:181
#1 0x4f4998 in view_child_destroy ../sway/tree/view.c:1131
#2 0x4f38fa in subsurface_handle_destroy ../sway/tree/view.c:946
#3 0x7fda50744892 in wlr_signal_emit_safe ../util/signal.c:29
#4 0x7fda5072f0dd in subsurface_destroy ../types/wlr_surface.c:649
#5 0x7fda507312c4 in subsurface_handle_surface_destroy ../types/wlr_surface.c:1094
#6 0x7fda50744892 in wlr_signal_emit_safe ../util/signal.c:29
#7 0x7fda5072f305 in surface_handle_resource_destroy ../types/wlr_surface.c:677
#8 0x7fda508180ce in destroy_resource (/lib64/libwayland-server.so.0+0xc0ce)
#9 0x7fda508187f2 in wl_client_destroy (/lib64/libwayland-server.so.0+0xc7f2)
#10 0x7fda50818e5f in wl_client_connection_data (/lib64/libwayland-server.so.0+0xce5f)
#11 0x7fda50818219 in wl_event_loop_dispatch (/lib64/libwayland-server.so.0+0xc219)
#12 0x7fda50818984 in wl_display_run (/lib64/libwayland-server.so.0+0xc984)
#13 0x43122c in server_run ../sway/server.c:254
#14 0x42f47c in main ../sway/main.c:433
#15 0x7fda503cab74 in __libc_start_main (/lib64/libc.so.6+0x27b74)
#16 0x40f6fd in _start (/opt/wayland/bin/sway+0x40f6fd)
0x6160000829a0 is located 288 bytes inside of 592-byte region [0x616000082880,0x616000082ad0)
freed by thread T0 here:
#0 0x7fda50f01a27 in free (/lib64/libasan.so.6+0xaea27)
#1 0x4532d8 in destroy ../sway/desktop/xdg_shell.c:262
#2 0x4ed17b in view_destroy ../sway/tree/view.c:67
#3 0x4ed300 in view_begin_destroy ../sway/tree/view.c:83
#4 0x454a3f in handle_destroy ../sway/desktop/xdg_shell.c:507
#5 0x7fda50744892 in wlr_signal_emit_safe ../util/signal.c:29
#6 0x7fda506e2c87 in reset_xdg_surface ../types/xdg_shell/wlr_xdg_surface.c:481
#7 0x7fda506e3018 in destroy_xdg_surface ../types/xdg_shell/wlr_xdg_surface.c:516
#8 0x7fda506dfbe5 in xdg_client_handle_resource_destroy ../types/xdg_shell/wlr_xdg_shell.c:71
#9 0x7fda508180ce in destroy_resource (/lib64/libwayland-server.so.0+0xc0ce)
previously allocated by thread T0 here:
#0 0x7fda50f01ed7 in calloc (/lib64/libasan.so.6+0xaeed7)
#1 0x454bc8 in handle_xdg_shell_surface ../sway/desktop/xdg_shell.c:528
#2 0x7fda50744892 in wlr_signal_emit_safe ../util/signal.c:29
#3 0x7fda506e2363 in handle_xdg_surface_commit ../types/xdg_shell/wlr_xdg_surface.c:378
#4 0x7fda5072e368 in surface_commit_state ../types/wlr_surface.c:455
#5 0x7fda5072e51d in surface_commit_pending ../types/wlr_surface.c:474
#6 0x7fda5072ea58 in surface_commit ../types/wlr_surface.c:542
#7 0x7fda4fb3ac03 in ffi_call_unix64 (/lib64/libffi.so.6+0x6c03)
Fixes#5168
If any error is encountered during execution of the first subcommand of
a freshly created bar configuration, parsing apparently is to be aborted
and the current bar config is freed. The pointer to that memory is left
dangling though, leading to a use-after-free on successive bar
subcommands. This quite reliably ends in a crash like so:
sway -c reproducer.config
00:00:00.083 [sway/config.c:865] Error on line 2 'foo bar': Unknown/invalid command 'foo' (s)
free(): double free detected in tcache 2
00:00:00.608 [swaynag/swaynag.c:451] failed to register with the wayland display
Aborted (core dumped)
Minimal reproducer config:
bar {
foo bar
position top
}
Other messages:
malloc(): unaligned fastbin chunk detected
double free or corruption (fasttop)
The invalid command has to be the first for a newly created bar config.
Removing the command or switching order so it's not the first one masks
the problem.
Prevent this from occuring by resetting the pointer to NULL after
freeing the memory.
Signed-off-by: Michael Weiser <michael.weiser@gmx.de>
the original behavior set fullscreen for all descendents of a container,
which causes issues when firefox is one of those children because it
sends its own set_fullscreen request in response to being fullscreened.
Can be useful to make sure a bugfix is included. In the future maybe
the wlroots version string could include a commit hash when built
from source, too.
When a tiling container is floated, the focus stack needs to be
appropraitely modified to return the container to its original
position in the tree upon floating disable, like i3.
Logic that obtains current DPMS state is put inside the handler.
sway_output from which the current DPMS state will be obtained is selected by the following logic:
* For '-' and '--' the focused output is used;
* For '*' error "Cannot apply toggle to all outputs" is reported;
* For everything else all_output_by_name_or_id() is used.
Fixes#5929.
When an application inhibited idle, a view pointer was stored and a
destroy listener was registered to the wlr inhibitor. As the wlr
inhibitor lives longer than the view, this lead to a dangling view
pointer between view unmap and inhibitor destroy.
Store a pointer to the wlr inhibitor instead of to the view, and look up
the view when needed, which may at any point be NULL. This also allows
for an inhibitor to remain functional if a surface is re-mapped.
container_at checks if the position provided matches the currently
focused container with view_container_at as a fast path.
view_container_at checks using the main container geometry, which
includes the titlebar and border area. If a tabbed container is focused,
then positions over unfocused tabs are incorrectly reported as belonging
to the focused container, breaking focus on click.
Add view_container_content_at for use in the focused container fast path
which only tests container content area, and fall back to full workspace
scans for border and titlebar areas.
Closes: https://github.com/swaywm/sway/issues/6074
container_at would maintain the current focus as long as a position was
over one of the container view's surfaces. If an oversized surface was
being clipped, this lead to weird focus behavior.
Instead, use view_container_at for this test, which intersects the
container box before looking at surfaces.
If a surface is associated with a sway container, we limit the
destination box to the container dimensions.
Floating views and popups are exempt from this clipping.
Previously, the special case handling of scratchpad and unmark commands
was (probably accidentally) limited to criteria directly handled in the
execute_command function. This would exclude: 1. for_window criteria, as
these are handled externally for views and 2. and mouse bindings which
select target the node currently under the mouse cursor.
As a concrete example `for_window [app_id="foobar"] move scratchpad,
scratchpad show` would show (or hide due to the toggling functionality)
another window from the scratchpad, instead of showing the window with
app_id "foobar".
This commit replaces the "using_criteria" flag with "node_overridden"
with the more general notion of signifying that the node (and
container/workspace) in the current command handler context of the sway
config is not defined by the currently focused node, but instead
overridden by other means, i.e., criteria or mouse position.
When issuing a focus command on a specific container, users expect to
proceed it even if is hidden by a fullscreen window.
This matches the behavior of i3.
In e0a94bee8d, it was believed that if the
container is being rendered, it must have an output.
This turned out not to be the case. When rendering a container, all its
children are rendered, even if the children is positioned off screen and
thus not having any output. This is the cause of the crash in #6061.
This commit introduces a null-check, which fixes#6061.
Before this commit, when an output had its scale dynamically changed,
Sway would not load a cursor theme with the new scale. This results
in stale cursor images when moving the cursor into an area controlled
by the compositor, like the background or resize areas.
To reproduce:
- Using IPC, set an output scale to a value that isn't currently used
- Move the cursor into a compositor-controlled area
- The cursor will not change
As of 66343839b1, sway now uses a
libdrm header. Add this dependency to the build system so headers from
it can be used on systems where pkg-config is required to find them.
On server request, we need to send configure events to inform the client
of the new intended size. If the client changes size itself, sending a
configure event will only cause problems.
Use transaction_commit_dirty_client to distinguish between the two
transaction causes.
Currently, various floating-point expressions involving
the coordinates of borders, titlebars and content surfaces
are directly assigned to integers, and so they are rounded
towards zero.
This results in off-by-one distances between these elements
when the signs of their coordinates differ.
Fixed by wrapping these expressions with a call to
floor before the assignment.
view_child_init was calling view_init_subsurfaces, which did not set the
parent attribute for the subchildren. This lead to the subchildren
acting as standalone children. If the parent was an xdg_popup, this
would make the subchild unaware of the popup position.
Introduce view_child_init_subsurfaces for view_child_init to use
instead.
Closes: https://github.com/swaywm/sway/issues/6038
The subchildren lose their parent association at this point, so they
will not be able to see that the parent is unmapped.
Instead, just set the subchildren to be unmapped directly.
When a container straddles multiple outputs, the title bar is only rendered
at the scale of the "effective" output. If the title bar straddles onto
another output with a different scale factor, it was drawn at the wrong size.
In this commit, we take into consideration the scale the title was rendered
at and scale it accordingly so that it appears at the right size on the other
outputs.
This fixes#6054.
To reproduce:
- Open a floating window and a popup that hangs over the bottom or right
- Move the window in the direction of the popup overhang
- The previous position of the popup is damaged, not the new one
