platform/upstream/kmscon.git
11 years agoseat: always activate fake-VTs during startup
David Herrmann [Sun, 9 Dec 2012 13:19:17 +0000 (14:19 +0100)]
seat: always activate fake-VTs during startup

Instead of relying on --switchvt, we should always activate fake-VTs
during startup. Otherwise, we might end up with dead seats as there is no
way to wake up kmscon later.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agokmscon: remove broken vt_eloop concept
David Herrmann [Sun, 9 Dec 2012 13:14:41 +0000 (14:14 +0100)]
kmscon: remove broken vt_eloop concept

We tried to avoid running unused code during shutdown by moving VT
handling into a separate eloop and run it exclusively. However, with
delayed VT switching, we need to run other event-handlers to make the VT
switch work. Therefore, run all of this during shutdown and remove the
vt_eloop event-loop.

This means that we might create new seats, displays, etc. during shutdown.
This is valid and doesn't harm kmscon, but we want to avoid doing stuff
that is undone few miliseconds later. Therefore, we set a "exiting" flag
and check that before initiating most of the big handlers.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: start hotplugged seats if --switchvt is given
David Herrmann [Sun, 9 Dec 2012 12:50:38 +0000 (13:50 +0100)]
seat: start hotplugged seats if --switchvt is given

We currently only start seats (switch to their VT) during startup but we
should also do that for hotplugged seats. Therefore, move that logic into
kmscon_seat and perform the startup after the seat is fully set up.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agokmscon: remove unused 'out' arguments
David Herrmann [Sun, 9 Dec 2012 12:44:52 +0000 (13:44 +0100)]
kmscon: remove unused 'out' arguments

We don't use the newly created objects so no need to return them.
Furthermore, we want the functions to already perform actions on the
objects so there is no guarantee the objects are still alive after
function return. Thus, simply return nothing.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agokmscon: move video-device uterm-monitor handling
David Herrmann [Sun, 9 Dec 2012 12:41:28 +0000 (13:41 +0100)]
kmscon: move video-device uterm-monitor handling

Same as previous commit: Move uterm-monitor device-data handling into the
real device create/destroy functions to allow dynamic device removal.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agokmscon: set uterm-seat data in seat-creation
David Herrmann [Sun, 9 Dec 2012 12:37:39 +0000 (13:37 +0100)]
kmscon: set uterm-seat data in seat-creation

If we want to be able to destroy seats from anywhere, not just
uterm-monitor callbacks, we need to be able to reset the monitor-event
seat-data so we no longer access the seat memory.
Hence, move the seat-data set/reset logic into the actual seat
creation/destroy functions.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: pass allowed VTs from caller
David Herrmann [Sun, 9 Dec 2012 12:26:45 +0000 (13:26 +0100)]
seat: pass allowed VTs from caller

We should handle everything --listen related in kmscon_main so pass in the
allowed-VTs mask from the caller.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoAdd manpage infrastructure and kmscon.1 manpage
David Herrmann [Sun, 9 Dec 2012 12:18:56 +0000 (13:18 +0100)]
Add manpage infrastructure and kmscon.1 manpage

Add a very basic man-page infrastructure so we can easily add new
docbook-style manpages.

Also add kmscon.1 man-page with a basic overview.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agouterm: fix compilation of builds where fbdev is the only output
James Buren [Thu, 6 Dec 2012 20:31:17 +0000 (14:31 -0600)]
uterm: fix compilation of builds where fbdev is the only output

Declare helpers as "static inline" as they might get included but not
used.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agocdev: change TODO item to check for CAP_KILL
David Herrmann [Thu, 6 Dec 2012 14:21:23 +0000 (15:21 +0100)]
cdev: change TODO item to check for CAP_KILL

CAP_KILL will be documented in kmscon manuals, but we should actually
check for the capability during startup.

However, this requires linking with libcap or libcap-ng and we currently
want to avoid that for such a simple task.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agokmscon: ignore SIGPIPE
David Herrmann [Thu, 6 Dec 2012 13:49:56 +0000 (14:49 +0100)]
kmscon: ignore SIGPIPE

Ignore SIGPIPE globally. We handle HUP correctly everywhere and this
signal is just annoying.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agocdev: do not hardcode TTY major
David Herrmann [Thu, 6 Dec 2012 13:45:52 +0000 (14:45 +0100)]
cdev: do not hardcode TTY major

Use TTY_MAJOR instead of hardcoding the literal 4.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agouterm: vt: use current VT if it is unused
David Herrmann [Thu, 6 Dec 2012 13:44:14 +0000 (14:44 +0100)]
uterm: vt: use current VT if it is unused

If we autoselect a VT and the current VT is unused, use it. This is
similar to the behavior of Weston and X-Server.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agokmscon: run only on VT-less seats in listen-mode
David Herrmann [Thu, 6 Dec 2012 13:18:57 +0000 (14:18 +0100)]
kmscon: run only on VT-less seats in listen-mode

If we run in listen mode, we are supposed to take over a seat. If a seat
supports VTs (like kernel VTs or kmscon cdev fake VTs) we assume that
there is some manager for these VTs (the one who created them). Therefore,
there is no need to run kmscon in listen mode on these seats. Instead, you
should run kmscon in default mode on these seats.

We enforce this limitation because if the VT-master on those seats dies
and causes a HUP, they have no way to notify us when they startup again.
Therefore, this kind of setup is broken. Furthermore, no-one would every
want such setups.

Instead use the startup mechanism of the VT/seat-manager to start kmscon
in default mode on those seats.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agokmscon: prevent --no-listen in combination with --cdev-session
David Herrmann [Thu, 6 Dec 2012 13:03:54 +0000 (14:03 +0100)]
kmscon: prevent --no-listen in combination with --cdev-session

cdev sessions should only be used in listen-mode to avoid stacking cdev
sessions.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agokmscon: introduce new --listen mode
David Herrmann [Thu, 6 Dec 2012 12:55:00 +0000 (13:55 +0100)]
kmscon: introduce new --listen mode

Instead of configuring everything for every kind of situation, we now
split kmscon into two modes: default-mode and listen-mode

In default-mode we run on the given seats once until we encounter a HUP or
until we are closed. It's a perfect replacement for agetty.

In listen-mode, we run only on non-VT seats and provide the full kmscon
functionality. We run as system daemon and wait for new seats and close
seats if they are destroyed. We simply provide a full system-console on
all seats.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agokmscon: replace gpu selection modes with new --gpu option
David Herrmann [Thu, 6 Dec 2012 10:36:46 +0000 (11:36 +0100)]
kmscon: replace gpu selection modes with new --gpu option

Instead of using one option for each selection mode, we now provide a new
option --gpus. It takes as argument the selection mode to use.

We also remove the --video-devices option as it doesn't fit into the big
picture. Instead, we should provide a blacklist and users would then use
--gpu=all --blacklist-video=<blacklist> instead.

However, blacklisting is also required for input devices and until we
figure it out, we will provide no blacklisting at all.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoMove .pc files to ./docs/pc/
David Herrmann [Wed, 5 Dec 2012 17:25:05 +0000 (18:25 +0100)]
Move .pc files to ./docs/pc/

We want only C-source files in ./src and the documents ./docs directory is
meant to contain any other sources.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agouterm: vt: fix opening new VTs
David Herrmann [Sun, 2 Dec 2012 22:27:00 +0000 (23:27 +0100)]
uterm: vt: fix opening new VTs

If no VT name is explicitly given, we need to use VT_OPENQRY to find one.
This used to work until we switched to supporting fake-VTs. Fix this again
to work like it used to.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: fix rescheduling sessions if none is active
David Herrmann [Sun, 2 Dec 2012 14:24:13 +0000 (15:24 +0100)]
seat: fix rescheduling sessions if none is active

If no session is active, we must not choose the scheduled session as
start-place. Otherwise, we will not schedule any session because the
start-session isn't used in the iteration.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agopty: use the new child-reaper sources
David Herrmann [Sun, 2 Dec 2012 14:08:07 +0000 (15:08 +0100)]
pty: use the new child-reaper sources

Instead of registering for SIGCHLD we should use the child-reaper sources
to avoid nasty race-conditions.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoeloop: add child-reaper event sources
David Herrmann [Sun, 2 Dec 2012 14:06:00 +0000 (15:06 +0100)]
eloop: add child-reaper event sources

If multiple childs die simultaneously, only a single SIGCHLD is queued.
Only after our process got scheduled and we unqueued the SIGCHLD, a next
signal can be queued. Therefore, using SIGCHLD as indicator that a
specific child died does not work reliably.

Instead, we must use waitpid() to wait for childs. Hence, we introduce a
new event-source which does that. It reports the PID/status combination
for each child to the registered callbacks.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoterminal: use "opened" logic
David Herrmann [Sun, 2 Dec 2012 13:59:12 +0000 (14:59 +0100)]
terminal: use "opened" logic

We already keep the open state as "opened" so we should use it instead of
always pushing the work down to the pty layer.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoeloop: fix memory loop when signal registration fails
David Herrmann [Sun, 2 Dec 2012 13:57:43 +0000 (14:57 +0100)]
eloop: fix memory loop when signal registration fails

If the signal-registration fails, we need to destroy it again. Otherwise,
it will keep a reference to the current eloop-object and hence it will
never get freed.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoshl: hook: allow callbacks to destroy the hook
David Herrmann [Sun, 2 Dec 2012 12:56:31 +0000 (13:56 +0100)]
shl: hook: allow callbacks to destroy the hook

If we currently dispatch a callback and they unregister themself and then
destroy the hook, we currently do not free the hook as we would free
memory that is used in the callstack.
However, it is a valid use-case for hook-callbacks to destroy the hook.
Therefore, we set a "dead" flag and destroy the hook after all callbacks
are done.

This fixes some memory-leaks inside the shared-signal callbacks of eloop.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agopty: simplify pty_close()
David Herrmann [Sun, 2 Dec 2012 12:42:15 +0000 (13:42 +0100)]
pty: simplify pty_close()

The old logic is no longer needed as we use edge-triggered fd notification
now. Therefore, simplify the helpers to make them more readable.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agotext: gltex: free atlas during destruction
David Herrmann [Sun, 2 Dec 2012 12:40:53 +0000 (13:40 +0100)]
text: gltex: free atlas during destruction

If we remove an atlas, we have to correctly free its memory. Otherwise, we
will loose the memory.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agopty: free seat argument on destruction
David Herrmann [Sun, 2 Dec 2012 12:40:16 +0000 (13:40 +0100)]
pty: free seat argument on destruction

We didn't correctly free this parameter even though we dup'ed it earlier.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: unref session after unregistration
David Herrmann [Sun, 2 Dec 2012 12:39:00 +0000 (13:39 +0100)]
seat: unref session after unregistration

Session registration does not provide a reference to the caller. Instead,
the seat itself owns the reference. Hence, we must clear this reference
during unregistration, otherwise we will never free the session structure.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: forward HUP to caller
David Herrmann [Sun, 2 Dec 2012 11:42:35 +0000 (12:42 +0100)]
seat: forward HUP to caller

We now forward VT_HUP to the caller so they can react on it. Default
behavior is to shutdown this seat. In single-seat mode we also close the
application.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: protect kmscon_seat_remove_display() correctly
David Herrmann [Sun, 2 Dec 2012 11:42:07 +0000 (12:42 +0100)]
seat: protect kmscon_seat_remove_display() correctly

We have to check for NULL pointers like all the other public entries do.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoconf: improve --vt warning by allowing --seats=,all
David Herrmann [Sun, 2 Dec 2012 11:29:18 +0000 (12:29 +0100)]
conf: improve --vt warning by allowing --seats=,all

If a seat is called "all" and we want to use it as single seat, we
currently have to pass "--seats=all," or "--seats=,all" etc. so it is not
detected as --seats=all.
However, our warning when --vt is used with it does not respect this.
Improve this warning to detect this case and correctly ignore it then.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoshl: misc: add shl_string_list_count() helper
David Herrmann [Sun, 2 Dec 2012 11:28:48 +0000 (12:28 +0100)]
shl: misc: add shl_string_list_count() helper

This helper counts the entries in a string-list with an option to ignore
empty elements.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: schedule dummy session if no other session is available
David Herrmann [Sun, 2 Dec 2012 11:07:06 +0000 (12:07 +0100)]
seat: schedule dummy session if no other session is available

In special situations (like forcing deactivation of a background session)
we might end up with no real registered session and no current session. In
these situations, we should allow the session-scheduler to schedule dummy
sessions.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: allow asynchronous command scheduling
David Herrmann [Sun, 2 Dec 2012 11:05:18 +0000 (12:05 +0100)]
seat: allow asynchronous command scheduling

When an asynchronous session-deactivation is finished, we need to know
which task to do now. As we cannot let the caller decide, because we are
asynchronous, we have to remember the task.

Three tasks are defined now but this can be extended. We reset the task
after the current session got deactivated. This guarantees that we never
have pending schedules but always return to a sane default even if we
missed a notification.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agouterm: vt: handle HUP on controlling TTY
David Herrmann [Sun, 2 Dec 2012 11:02:06 +0000 (12:02 +0100)]
uterm: vt: handle HUP on controlling TTY

If the controlling TTY (in our case the VT) signals us a HUP, we should
consider it closed and forward that to the caller. This does normally not
happen on real VTs, however, on fake-VTs this happens if the parent goes
away.

We forward the HUP to the controlling subsystem which then has to deal
with it. But more importantly, we disable the fd so we don't hog the CPU
by unconditionally flushing the queue.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agocdev: introduce client 'dead' state
David Herrmann [Sun, 2 Dec 2012 10:03:40 +0000 (11:03 +0100)]
cdev: introduce client 'dead' state

Files have the disadvantage that the kernel (or in this case 'we') cannot
force a client to close it. Instead, we have to mark it as 'dead' and wait
until the client closes the file theirself.
While marked as 'dead' we always return EPIPE or signal EPOLLHUP so the
client will notice that the file is dead.

This fixes a bug where when we forced an unregistration of a cdev session,
the cdev_client went away too early and the user still had the file open.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agocdev: fix copying pollhandles correctly
David Herrmann [Sun, 2 Dec 2012 10:02:17 +0000 (11:02 +0100)]
cdev: fix copying pollhandles correctly

The pollhandle is solely implemented in libfuse and does not correspond to
some kernel-intern object. Instead, the kernel notifies us when it has
poll-waiters via a flag. If that flag is set we should notify the kernel
whenever the poll-flags change.

Therefore, we can safely copy the pollhandle all the time and delete our
temporary handles.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agouterm: vt: add uterm_vt_get_type() helper
David Herrmann [Sun, 2 Dec 2012 09:19:43 +0000 (10:19 +0100)]
uterm: vt: add uterm_vt_get_type() helper

This helper returns the VT type. This will allow external users to see
whether we have real VTs or whether we don't.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agouterm: vt: set K_RAW before trying K_OFF
David Herrmann [Sun, 2 Dec 2012 09:18:15 +0000 (10:18 +0100)]
uterm: vt: set K_RAW before trying K_OFF

K_OFF is not available on older kernels and it is currently discussed
whether it should be replaced on newer kernels. Therefore, set K_RAW
before trying K_OFF and then ignore errors if K_OFF cannot be set.

Reported-by: Ran Benita <ran234@gmail.com>
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agocdev: fix delayed deactivation to update internal state
David Herrmann [Sun, 2 Dec 2012 09:16:24 +0000 (10:16 +0100)]
cdev: fix delayed deactivation to update internal state

We need to set "active" to false during delayed deactivation, as
otherwise we have inconsistent state in respect to the seat subsystem.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agotsm: vte: ignore DEC SET/RESET 12 mode
David Herrmann [Sun, 2 Dec 2012 09:15:40 +0000 (10:15 +0100)]
tsm: vte: ignore DEC SET/RESET 12 mode

This mode is used to start/stop blinking cursors. As vim uses this
heavily, we should simply ignore it to avoid printing useless debug
messages.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: ignore keyboard shortcuts in single-session mode
David Herrmann [Sat, 1 Dec 2012 22:45:55 +0000 (23:45 +0100)]
seat: ignore keyboard shortcuts in single-session mode

Instead of not handling the keyboard shortcuts we should ignore them but
mark them as handled. Otherwise, we might get them from the parent session
and they will end up cluttering the terminal log.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agocdev: implement VT_WAITACTIVE
David Herrmann [Sat, 1 Dec 2012 18:06:38 +0000 (19:06 +0100)]
cdev: implement VT_WAITACTIVE

The VT API provides a VT_WAITACTIVE ioctl which waits until the VT got
activated. We use the same technique as with blocking reads() to implement
this without threads.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoterminal: fix unregistering display callbacks when removing screens
David Herrmann [Sat, 1 Dec 2012 17:22:38 +0000 (18:22 +0100)]
terminal: fix unregistering display callbacks when removing screens

We must unregister the display callback when removing a screen. Otherwise,
we keep writing to the screen memory which becomes invalid then.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: change default keyboard shortcuts to ctrl+logo
David Herrmann [Sat, 1 Dec 2012 16:39:38 +0000 (17:39 +0100)]
seat: change default keyboard shortcuts to ctrl+logo

ctrl+alt is already used by many other applications so lets use some other
unused shortcut modifiers.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoconf: introduce --configdir parameter
David Herrmann [Sat, 1 Dec 2012 15:21:33 +0000 (16:21 +0100)]
conf: introduce --configdir parameter

The --configdir parameter allows specifying another configuration
directory than the default /etc/kmscon.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: limit multi-session mode to --multi-session
David Herrmann [Sat, 1 Dec 2012 15:09:14 +0000 (16:09 +0100)]
seat: limit multi-session mode to --multi-session

If kmscon is started without the --multi-session option, we no longer
provide multi-session support (now the default). This is for better
backwards compatibility to agetty and friends.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoconf: print warning if --vt is used wrong
David Herrmann [Sat, 1 Dec 2012 14:46:42 +0000 (15:46 +0100)]
conf: print warning if --vt is used wrong

--vt should only be used when kmscon runs on a single seat. Otherwise,
this VT will be forced on all seats and each kmscon instance _might_
collide.
If this is done with care, the --vt/--seats combination can work. However,
the use-cases are negligible and we should rather warn the ordinary user
not to do it.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agouterm: vt: use real VTs on all seats if available
David Herrmann [Sat, 1 Dec 2012 14:25:37 +0000 (15:25 +0100)]
uterm: vt: use real VTs on all seats if available

If there is already a kmscon instance running that provides fake-VTs, then
lets use it. This allows stacking kmscon processes and is very useful for
debugging.

Also other applications that use uterm can now run inside of kmscon
without any other special configuration.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: register cdev session only when --cdev is given
David Herrmann [Sat, 1 Dec 2012 14:05:07 +0000 (15:05 +0100)]
seat: register cdev session only when --cdev is given

We cannot have multiple instances of kmscon running with cdev emulation,
so disable it by default. A kmscon-system-daemon now needs --cdev if you
want fake VTs to work.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoconf: fix whitespace offset for -s
David Herrmann [Sat, 1 Dec 2012 13:50:53 +0000 (14:50 +0100)]
conf: fix whitespace offset for -s

Align everything to the same offset.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agopty: downgrade error to debug message on I/O errors
David Herrmann [Sat, 1 Dec 2012 12:15:01 +0000 (13:15 +0100)]
pty: downgrade error to debug message on I/O errors

I/O errors aren't fatal or important on PTYs so ignore them and mark it as
debug message.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agocdev: fix typo in warning
David Herrmann [Sat, 1 Dec 2012 12:11:39 +0000 (13:11 +0100)]
cdev: fix typo in warning

Obvious typo. Got those words swapped...

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agouterm: clear displays during sleep and wakeup
David Herrmann [Wed, 28 Nov 2012 21:54:46 +0000 (22:54 +0100)]
uterm: clear displays during sleep and wakeup

We shouldn't leave the content on the framebuffers when leaving kmscon.
This might confuse the user. Furthermore, we shouldn't reload the old
framebuffer during wakeup as this frame can be quite old and we see a
short flickering during wakeup which isn't wanted, either.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: set DPMS to ON before broadcasting the device
David Herrmann [Wed, 28 Nov 2012 21:21:33 +0000 (22:21 +0100)]
seat: set DPMS to ON before broadcasting the device

We cannot modeset a device while DPMS is set to OFF so we should do this
before broadcasting the device. Otherwise, the callback might not be able
to use the display.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agomain: correctly remove all displays on uterm-video shutdown
David Herrmann [Wed, 28 Nov 2012 21:16:07 +0000 (22:16 +0100)]
main: correctly remove all displays on uterm-video shutdown

If we close an uterm-video object, we must correctly remove all displays
first.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoterminal: check swapping state during activation
David Herrmann [Wed, 28 Nov 2012 21:14:05 +0000 (22:14 +0100)]
terminal: check swapping state during activation

If a terminal gets activated while a swap-buffer operation is still in
progress from the previous session, we must not directly reschedule it.
Instead, we wait for the page-flip event and continue then.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: activate displays only when being in foreground
David Herrmann [Wed, 28 Nov 2012 20:55:50 +0000 (21:55 +0100)]
seat: activate displays only when being in foreground

We shouldn't activate displays when we are active but in background. In
these moments, the uterm-video device might not be initialized, yet.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoterminal: remove now unused --fps option
David Herrmann [Wed, 28 Nov 2012 19:57:36 +0000 (20:57 +0100)]
terminal: remove now unused --fps option

We no longer have a framerate-control option so remove it.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoterminal: sync redraw events with vertical blanks
David Herrmann [Wed, 28 Nov 2012 19:51:30 +0000 (20:51 +0100)]
terminal: sync redraw events with vertical blanks

We used an FPS timer to control redraw events but this caused some
unwanted artifacts. Instead, we should always wait for the page-flip event
before redrawing the screen.

This can still be optimized (for instance via triple-buffering) but it
works very nice now.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoterminal: keep terminal reference in each screen
David Herrmann [Wed, 28 Nov 2012 18:31:09 +0000 (19:31 +0100)]
terminal: keep terminal reference in each screen

We need to have a terminal reference in future screen callbacks so keep
one from the beginning.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agopty: fix ignoring HUP on pty-master files
David Herrmann [Wed, 28 Nov 2012 16:47:31 +0000 (17:47 +0100)]
pty: fix ignoring HUP on pty-master files

A HUP is reported whenever there is no client listening on the other side
of the PTY. As /bin/login and friends call vhangup() during setup, there
is a short time-span when we might get a HUP.
Hence, we need to ignore HUPs and solely track clients by PID. To avoid
hogging the CPU while the TTY is hungup, we change the event-mode to
edge-triggered.

This fixes a bug where we tried starting /bin/login several times because
we always ran into this race-condition and directly closed the TTY again.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoeloop: allow edge-triggered operation
David Herrmann [Wed, 28 Nov 2012 16:46:18 +0000 (17:46 +0100)]
eloop: allow edge-triggered operation

A new EV_ET flag allows ev_fd sources to operate in edge-triggered instead
of level-triggered mode. See epoll(7) for more information.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agopty: remove unused "master" argument
David Herrmann [Tue, 27 Nov 2012 18:02:48 +0000 (19:02 +0100)]
pty: remove unused "master" argument

We do not use the master pty fd when spawning the child so skip it.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoeloop: fix not disabling sources on EV_HUP
David Herrmann [Tue, 27 Nov 2012 18:00:57 +0000 (19:00 +0100)]
eloop: fix not disabling sources on EV_HUP

We used to disable event sources on EV_HUP to avoid draining the
event-queue for non-edge-triggered sources. However, this turns out to be
not working with special files so we disable it again.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agouterm: video: add uterm_display_is_swapping() helper
David Herrmann [Mon, 26 Nov 2012 16:19:45 +0000 (17:19 +0100)]
uterm: video: add uterm_display_is_swapping() helper

This new helper can be used to check whether a display has a pending
vsync'ed buffer-flip scheduled.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agouterm: video: remove uterm_screen API
David Herrmann [Mon, 26 Nov 2012 16:11:07 +0000 (17:11 +0100)]
uterm: video: remove uterm_screen API

The uterm_screen API is no longer used internally and we do not plan to
extend it. So we can use uterm_display as a replacement to make the API
thinner and easier to use.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agotest_output: use uterm_display directly
David Herrmann [Mon, 26 Nov 2012 16:10:44 +0000 (17:10 +0100)]
test_output: use uterm_display directly

uterm_screen is scheduled for removal so use the display API directly.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agotext: use uterm_display directly
David Herrmann [Mon, 26 Nov 2012 15:50:39 +0000 (16:50 +0100)]
text: use uterm_display directly

Instead of using the uterm-screen indirection, we now directly access
uterm-display objects. We do not really intend to use virtual screens with
kmscon so there is no need to make this more complex.

I don't think consoles should every provide this feature. Instead, you
should use real compositors for such tasks like Wayland+wlterm. kmscon,
however, is rather an emergency tool or a safe backup than a fancy
daily-use-console.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoui: remove old unused UI subsystem
David Herrmann [Tue, 20 Nov 2012 13:30:54 +0000 (14:30 +0100)]
ui: remove old unused UI subsystem

The UI subsystem was replaced by the seat subsystem and is no longer used.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agotsm: handle ISO_Left_Tab
David Herrmann [Mon, 19 Nov 2012 15:26:38 +0000 (16:26 +0100)]
tsm: handle ISO_Left_Tab

If Shift+Tab is pressed, most keyboard layouts map this to ISO_Left_Tab
which itself should be converted to \e[Z.

Reported-by: Vladimir Kravets
Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agocdev: implement fake-VT-switching
David Herrmann [Mon, 19 Nov 2012 15:14:38 +0000 (16:14 +0100)]
cdev: implement fake-VT-switching

This implements VT switching for cdev sessions. This allows to run
external applications as sessions inside of kmscon.

This should only be used for backwards compatibility. There is really no
reason to develop new applications to do this. Instead, use
system-compositors which make all the work for you!

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: add --grab-session-dummy option
David Herrmann [Mon, 19 Nov 2012 15:11:34 +0000 (16:11 +0100)]
seat: add --grab-session-dummy option

This option specifies a shortcut that can be used to schedule the dummy
session on the given seat.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: add helpers to schedule sessions from the outside
David Herrmann [Mon, 19 Nov 2012 15:10:01 +0000 (16:10 +0100)]
seat: add helpers to schedule sessions from the outside

Two new helpers to schedule a specific session or a session given by a
hint-number. These can be used by other subsystems to schedule
their/others sessions.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: remove "manual input" logic
David Herrmann [Mon, 19 Nov 2012 15:07:34 +0000 (16:07 +0100)]
seat: remove "manual input" logic

This really doesn't make sense. We no longer want the applications to be
in charge of VT switches so always force the VT-master to be in control of
the VT-switching logic.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: allow scheduling dummy session
David Herrmann [Mon, 19 Nov 2012 15:03:57 +0000 (16:03 +0100)]
seat: allow scheduling dummy session

We always considered the dummy session to be no real session. However,
under special circumstances it might be desirable to schedule the dummy
session (very important for debugging). Therefore, change the logic to
allow scheduling the dummy session.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agouterm: vt: fix input-device sleep-control
David Herrmann [Mon, 19 Nov 2012 14:57:35 +0000 (15:57 +0100)]
uterm: vt: fix input-device sleep-control

We need to correctly put input devices asleep only if we really disable
the VT. If our callbacks fail and prevent the VT-switch, we must keep the
input devices awake. Otherwise, we will lose control over the
sleep-states.

Also correctly put the device asleep during shutdown.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoshl: array: use "const" for source arguments
David Herrmann [Mon, 19 Nov 2012 12:54:17 +0000 (13:54 +0100)]
shl: array: use "const" for source arguments

Use "const" even for void pointers as gcc will complain otherwise. Source
arguments are read-only, anyway, so this doesn't break anything.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: be more verbose on asynchronous session switching
David Herrmann [Mon, 19 Nov 2012 12:18:53 +0000 (13:18 +0100)]
seat: be more verbose on asynchronous session switching

Print debug messages when doing asynchronous session switches and handle
EINPROGRESS as special error code.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: fix delayed VT switches
David Herrmann [Mon, 19 Nov 2012 12:13:01 +0000 (13:13 +0100)]
seat: fix delayed VT switches

If we switch VT and the current session cannot be put asleep, and later it
notifies us that it went asleep, we should then switch VT and not
reschedule the sessions.

We do this by remembering whether a VT switch is pending. The new
uterm-retry infrastructure allows us to easily repeat VT switches then.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: fix calling parent-callbacks on forced sleeping
David Herrmann [Mon, 19 Nov 2012 12:12:10 +0000 (13:12 +0100)]
seat: fix calling parent-callbacks on forced sleeping

If we force a seat to go asleep, we should _always_ notify the parent
about this. Otherwise, we might get unexpected results.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: allow sessions to switch between background/foreground
David Herrmann [Mon, 19 Nov 2012 12:10:51 +0000 (13:10 +0100)]
seat: allow sessions to switch between background/foreground

Two new helpers allow sessions to mark them as background and foreground
sessions. This will immediately switch them into the new state.

The internal infrastructure was already available, but we currently always
used foreground sessions.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: allow sessions to control VT-switches
David Herrmann [Mon, 19 Nov 2012 12:06:43 +0000 (13:06 +0100)]
seat: allow sessions to control VT-switches

A new helper allows sessions to manually handly input events so the core
does no longer perform session/VT switches on their behalf.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agouterm: vt: implement delayed VT-switches
David Herrmann [Mon, 19 Nov 2012 12:04:15 +0000 (13:04 +0100)]
uterm: vt: implement delayed VT-switches

If we cannot immediately perform a VT switch away from our application, we
might want to retry this shortly after. Therefore, we can now call
uterm_vt_retry() which then performs any outstanding VT-switches.

To avoid unexpected behavior, we limit this to a 2-3 second delay.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agouterm: vt: do nothing if VT-switch targets our VT
David Herrmann [Mon, 19 Nov 2012 12:01:57 +0000 (13:01 +0100)]
uterm: vt: do nothing if VT-switch targets our VT

If the user presses some keys to switch VTs and the target VT is our VT
(which is already active in this case), we shouldn't do anything. Calling
VT_ACTIVATE doesn't particularly hurt, but it doesn't get us anything
here so skip it.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agouterm: vt: do nothing if target-VT is already active during deactivation
David Herrmann [Mon, 19 Nov 2012 12:00:38 +0000 (13:00 +0100)]
uterm: vt: do nothing if target-VT is already active during deactivation

If we deactivate our VT and the target-VT is already active (probably
because we opened the _current_ VT during uterm-vt creation), we shouldn't
do anything but simply close the VT.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: fix forwarding errors in seat_pause()
David Herrmann [Sat, 17 Nov 2012 19:25:29 +0000 (20:25 +0100)]
seat: fix forwarding errors in seat_pause()

If we force a pause we need to forward the error code. Otherwise, we
cannot properly react on it and print warnings.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: allow calling go_foreground if already in foreground
David Herrmann [Sat, 17 Nov 2012 19:13:27 +0000 (20:13 +0100)]
seat: allow calling go_foreground if already in foreground

If we are asleep and in foreground, go_foreground() should return success
so we can safely unload asynchronous sessions. Otherwise, we wouldn't be
able to handle them properly while being inactive or we would need a whole
bunch of other checks.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agouterm: vt: do not check SIGUSR1/2 for SI_KERNEL
David Herrmann [Sat, 17 Nov 2012 18:45:44 +0000 (19:45 +0100)]
uterm: vt: do not check SIGUSR1/2 for SI_KERNEL

If we want to run as client under user-space fake-VT implementations like
kmscon-cdev, we need to be able to catch those signals from
userspace-daemons. Therefore, do not check whether the signals were
generated by the kernel.

This doesn't affect security in any way as non-priviledged processes
aren't allowed to send signals, anyway. Moreover, this increases
debugability of VT layers a lot as we can fake these signals now.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agobuild: fix wlterm wayland-dependency check
David Herrmann [Sat, 17 Nov 2012 18:42:57 +0000 (19:42 +0100)]
build: fix wlterm wayland-dependency check

The logic behind this two-layer check is bogus and doesn't work. Merge
both layers so we have a proper dependency-check again. This will make the
error messages less verbose but at least it works again.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: remove bogus debug statements
David Herrmann [Sun, 11 Nov 2012 11:17:11 +0000 (12:17 +0100)]
seat: remove bogus debug statements

These shouldn't be pushed into the public repository. Sorry.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agotest_vt: add --switchvt option
David Herrmann [Sun, 11 Nov 2012 10:39:38 +0000 (11:39 +0100)]
test_vt: add --switchvt option

This has the same effect as for kmscon but is disabled by default.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agotest_vt: add --vt=<path> option
David Herrmann [Sun, 11 Nov 2012 10:37:17 +0000 (11:37 +0100)]
test_vt: add --vt=<path> option

This new option allows specifying a path to the VT that should be used.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agotest_vt: remove trailing newline in log-message
David Herrmann [Sun, 11 Nov 2012 10:36:38 +0000 (11:36 +0100)]
test_vt: remove trailing newline in log-message

The log-subsystem adds these automatically so we shouldn't use them.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agotest_vt: use a dummy input-object
David Herrmann [Sun, 11 Nov 2012 10:35:15 +0000 (11:35 +0100)]
test_vt: use a dummy input-object

We cannot allocate VT objects without an associated input object.
Therefore, allocate a dummy device with no input devices assigned to it
and use it for the test-vt object.

Note that this makes switching from/to this VT via command-line
impossible.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agouterm: vt: fix not overwriting 'ret' in error paths
David Herrmann [Sun, 11 Nov 2012 10:33:19 +0000 (11:33 +0100)]
uterm: vt: fix not overwriting 'ret' in error paths

We cannot use 'ret' in error paths, otherwise, we might return 0. This
might then cause instabilities as the objects weren't initialized
correctly.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agouterm: vt: do not modify TTY settings
David Herrmann [Sat, 10 Nov 2012 17:11:16 +0000 (18:11 +0100)]
uterm: vt: do not modify TTY settings

We do not use the TTY for anything except VT functions. Hence, there is no
need to modify the TTY settings before using it.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: allow sessions to notify seat when deactivating
David Herrmann [Sat, 10 Nov 2012 15:57:04 +0000 (16:57 +0100)]
seat: allow sessions to notify seat when deactivating

If a session is deactivated asynchronously, it can now notify the seat
when it is done. This will then cause a re-schedule of the session list.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
11 years agoseat: implement delayed session switching
David Herrmann [Sat, 10 Nov 2012 15:46:58 +0000 (16:46 +0100)]
seat: implement delayed session switching

If we want to allow external programs to control sessions inside of
kmscon, we need to give them some time for session
activation/deactivation. Therefore, the whole session-management is
asynchronous now and allows the session code to return EINPROGRESS or
similar if they need more time.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>