Ray Strode [Tue, 23 Mar 2010 21:48:48 +0000 (17:48 -0400)]
[set-default-theme] Write plymouthd.conf instead of symlink
Now that the daemon looks for the default theme in configuration
files, we should make plymouth-set-default-theme write the
configuration files instead of doing symlinks.
Charlie Brej [Mon, 22 Mar 2010 23:39:32 +0000 (23:39 +0000)]
[script] Update default script to use the window X and Y offsets every time
Previously the code was assuming the windows were placed at 0,0. This might not
be the case and the window X and Y values should be used when trying to
position items relative to a window.
This change needs to be applied to all other scripts otherwise mutiple screen
setups may have unaligned elements. Updates scripts should be tested using
multi-head test systems or the x11 test renderer.
Charlie Brej [Mon, 22 Mar 2010 23:16:21 +0000 (23:16 +0000)]
[script] Get window value without an index returns area covered by all windows
Calls to Window.GetWidth/Height/X/Y without a window index now return the
values of the area covered by all windows. This is only the case if all the
windows are aligned (either by their centers, or to a corner).
This allows the theme designer to place an object knowing it will be seen on
all screens.
Charlie Brej [Mon, 22 Mar 2010 22:41:17 +0000 (22:41 +0000)]
[script] Return NULL when requesting width of nonexistent window
Minor bug, previously would return the index used rather than a NULL. Would
only cause problems when using a width request as a test of the presence of a
window.
Charlie Brej [Mon, 22 Mar 2010 22:26:47 +0000 (22:26 +0000)]
[script] Move windows to not start at negative coordinates
When multiple screens are found, the system will now arrange them so they are
all centered, and the top left corner of the largest screen is at 0,0.
No changes to any scripts are needed.
Ray Strode [Mon, 22 Mar 2010 04:13:35 +0000 (00:13 -0400)]
[main] Deactivate terminal on quit if necessary
If we're done with the VT plymouth was running on,
and plymouth wasn't running on the initial VT, we
should jump back to the initial VT and try to
clean up plymouth's VT.
Resetting the mode to text on every write means that if you're
using a text plugin and X starts, X's VT keeps getting reset back to
KD_TEXT since those plugins don't stop writing on deactivate (they
have no renderer).
There's no reason to set this mode here anyway; all paths to using
those plugins already do this.
[terminal] keep track of terminal raw/cooked state
Instead of setting the terminal to unbuffered (raw) mode on every
write, keep track of whether it's unbuffered or not at the points
we open and close the terminal.
Deactivate already takes care to set back into buffered mode;
otherwise we can end up resetting the terminal mode under X causing
Enter to send X SIGQUIT.
If we don't deactivate the renderer before hiding the splash, the
drm renderer may scan out the buffer contents to the fbcon buffer;
since we only hide the splash when dumping details or when
--retain-splash is *not* given to quit, this is exactly the
opposite of what we want.
The effect of not doing this is partial splash contents behind the
details in cases of error, or when using quit. This doesn't affect
plymouth quit --retain-splash.
One problem with the current deactivate/quit transition into X is that
the display manager will, if Plymouth was running, re-use the currently
active VT.
That only works if Plymouth was actually displaying a splash screen on
that VT. If --show-splash hasn't been called yet because we booted too
fast, we'll be on the wrong VT.
Add a request to ask whether the Plymouth VT is active; I've done it
this way so the answer defaults to "yes" for Fedora who use VT1.
The pseudo-code for transition is thus:
if plymouth is running (ping):
plymouth deactivate
if plymouth has active vt:
start X on current VT with -nr
if X starts ok:
plymouth quit --retain-splash
else if X fails:
plymouth quit
else if plymouth doesn't have active vt:
plymouth quit
start X as normal
else if plymouth isn't running:
start X as normal
Change the display_normal() function so that rather than being a no-op
if we already saved the state as normal, it restarts any animations and
redraws the views.
The only thing we now do if the state is not previously the same is
hide any prompt.
This allows this to be used to reanimate the plugin on reactivate.
Change the display_normal() function so that rather than being a no-op
if we already saved the state as normal, it restarts any animations and
redraws the views.
The only thing we now do if the state is not previously the same is
hide any prompt.
This allows this to be used to reanimate the plugin on reactivate.
[space-flares] resume animations on display_normal
Change the display_normal() function so that rather than being a no-op
if we already saved the state as normal, it restarts any animations and
redraws the views.
The only thing we now do if the state is not previously the same is
hide any prompt.
This allows this to be used to reanimate the plugin on reactivate.
[fade-throbber] resume animations on display_normal
Change the display_normal() function so that rather than being a no-op
if we already saved the state as normal, it restarts any animations and
redraws the views.
The only thing we now do if the state is not previously the same is
hide any prompt.
This allows this to be used to reanimate the plugin on reactivate.
Since deactivate uses on_boot_splash_idle, there's a good chance that
plugins will have stopped animating. Prod them to animate again by
calling update_display()
More for debugging and completeness than anything else, add a
"reactivate" command to the daemon that undoes the effects of
deactivate and continues the splash screen on its way.
Another possible use for this could be (for example) providing a
seamless shutdown experience.
A future commit will implement the client bits needed.
Currently deactivate is mostly like hide splash, except it deactivates
the renderer first and doesn't reset the VT to text mode and dump the
details plugin over top.
Unfortunately this means that the renderer is closed and freed, and in
the case of the Intel DRM renderer, closing the DRM file descriptor
means that the kernel frees the buffers and restores the fbcon buffer
on our screen - losing the smooth transition.
This now changes deactivate such that it leaves the boot splash open,
but in an inactivate state, with the DRM connection still open, etc.
now the fbcon contents are not restored.
We deliberately stop watching for keyboard input, detach any logging
session from the console, take the VT out of VT_PROCESS mode and
put it back into cooked mode, etc.
This means the X server can be started, and this state can be cleaned
up by calling plymouth quit with affecting X.
[main] call deactivate_splash() even when no splash
To allow deactivate_splash() to do more than affect the boot splash
plugin and renderer, call it from the deactivate path when there is
no boot splash screen as well.
[terminal] export functions to enable/disable VT watching
In order to deactivate without pulling everything, we need to be able
to take the terminal in and out of VT_PROCESS mode directly; so change
the two functions from static to exported.
Since we ignore --show-splash, it makes no sense to process
--hide-splash either; in theory this does nothing already because
we won't have a boot_splash in our state - but that changes with
future patches and it's worth being safe.
Ray Strode [Mon, 22 Mar 2010 02:59:09 +0000 (22:59 -0400)]
[main] Ensure second quit blocks until quit processed
When it comes to quitting, which must be carefully coordinated
with other subsystems during the boot process, we need to make
sure the client doesn't return early.
Ray Strode [Mon, 22 Mar 2010 02:59:09 +0000 (22:59 -0400)]
[main] Ensure second deactivate blocks until deactivation
When it comes to deactivation where the various steps of boot
have to be carefully synchronized, we need to make sure that
the client doesn't return early.
Likewise if plymouth quit is called when we're waiting for the boot
splash to become idle, we also end up asserting that there's not
already an idle trigger.
Fix it in the same way as deactive, ignore the second quit command
except for pulling its trigger.
If plymouth deactivate is called when we're waiting for the boot
splash to become idle, we end up asserting that there's not
already an idle trigger.
Fix it by checking for an existing deactive trigger, and if there
is, ignoring the new deactivate command (except for pulling its
trigger so it doesn't block).
[main] give quit command precedence over deactivate
In the cases where the boot splash plugin does not become idle
immediately, we go back into the main loop and can receive additional
commands.
Since quit and deactive both use this facility, one scenario is the
quit command arriving after the deactivate command, but before the
deactivate command has actually been run.
In that situation, we want to quit, not deactivate.
One of the main differences between Ubuntu's use of Plymouth and
Fedora's is that on Ubuntu we've tried to keep the X server on VT7
so that the historical documentation of Ctrl-Alt-F1 giving you a
text console is preserved.
This obviously means that for a smooth transition, Plymouth must also
run on VT7.
We discovered that although Plymouth does have code to attempt to deal
with VTs, none of it is quite right and there are many paths that don't
work unless Plymouth is run on VT1.
This patch set fixes our known problems with the VT handling making it
possible to run Plymouth on any VT of your choosing, with VT1 remaining
the default.
Add a command-line option to specify the TTY that plymouth should
use. This is mostly useful for debugging, for example you can put
plymouth onto a TTY not used by the X server; or if you're feeling
particularly sneaky, plymouth into an xterm.
Change the renderer so that it defaults to inactive, then when we
map to the device, activate the renderer by activating the VT;
unless the VT is already active in which case activate the
renderer directly.
Change the renderer so that it defaults to inactive, then when we
map to the device, activate the renderer by activating the VT;
unless the VT is already active in which case activate the
renderer directly.
[terminal,text,details] move activate vt into plugins
Move the responsiblity to activate the VT into the text and details
plugins; this not only matches the graphical renderers, but it also
ensures we activate the VT in all possible code paths.
Previously if we fell back to text.so because we couldn't activate
a renderer, this would not activate the VT.
[terminal] move terminal opening into renderers/plugins
Move the responsibility to actually open the terminal to the renderers
and text/details plugin, this allows the X11 renderer to not actually
open the terminal - and thus not crash the X server.
When we activate our VT, now we actually have it in VT_PROCESS not the
VT we started from, we get the proper signal so don't need an ioctl to
wait until its active.
If we were to leave our VT, we'd get the opposite signal as well and
we don't really care whether we actually get to the other VT, just
that we leave ours.
[terminal] don't keep track of active vt, just if vt is active
Trying to keep track of whatever VT is actually active is inherently
racy; instead just keep track of whether our VT is the active one.
Since we guarantee that's the VT in VT_PROCESS now, this is easy.
Rather simplifies the on_vt_changed functions in renderers too.
Now we can simply test whether a terminal is virtual or not, add
a guard in the mode and vt change functions that ensure they only
operate on virtual terminals.
We want to be able to query whether a terminal is an ordinary terminal
or a virtual terminal, rather than looking up the vt number and knowing
what the right numbers are, add a function that knows that.
Remove all of the references to ply_console_t from the code, now
we operate exclusively on the terminal object. In some places
this means switching from one to the other, but in many it just
means dropping the console object and using the terminal object
we were already passed.
This removes the separation in code of "console" functions and
"terminal" functions; this never really made sense, and doesn't
particularly map to the behaviour of Linux virtual terminals.
The three principle operations that Plymouth was using "console"
for were:
* changing the active VT
* notification of changes to the active VT through VT_PROCESS
* switching between text and graphics mode
And it was using the "foreground terminal" alias /dev/tty0 to do
this. While this is fine for the first of those, since any console
device will do, it's always wrong for the latter two which should
always be on the actual VT we want Plymouth to run from.
If running on tty7, only tty7 should be in VT_PROCESS mode (since
we want to know when we enter this VT and leave this VT), and
certainly only tty7 should be in graphics mode.
Since you can use that same tty to obtain the current active VT,
and switch VT, you don't need another; so the need for a separate
"console" functionality goes away.
Since script implements a set_keyboard function (the only plugin to
do so), the previous commit reveals a bug where set_keyboard is
called for the plugin but unset_keyboard isn't called if the plugin
fails to be loaded
The script plugin only works on pixel displays, however there wasn't
any check for this, so if a script-based theme was your default
Plymouth would not fallback to using the text plugin instead.
Richard Maw [Wed, 10 Mar 2010 20:28:09 +0000 (20:28 +0000)]
[script] Add SubString function to the script string library
Adds the SubString function which returns a string segment. The two paramiters
are the sub-string start and end indicies. Negative start and end values return
a NULL, as does start index being beyond the end index. Start being beyond the
end of the string returns an empty string.
The on_draw() function inside the script plugin isn't referenced
anywhere, but references a static function from script-lib-sprite.c;
if building without optimisation, it's possible that gcc won't elide
this code so will fail during linking.
[scripts] Don't hardcode LIBEXECDIR and DATADIR paths
The scripts hard-coded the paths for LIBEXECDIR and DATADIR, unless
passed as environment variables. Instead of doing this, which breaks
if plymouth is installed outside of /usr, set these derived from the
configure $libexecdir and $datadir variables just as we do for
pkg-config, etc.
Since we use so many variables, it makes more sense to generate these
scripts from config.status rather than having special Makefile rules
for them.
Add a --pid-file option to plymouthd that will cause the daemon's
pid to be written to the named file. Useful to avoid grovelling
through ps output to find it again.
[client] Install libply-boot-client library and headers
When communicating with Plymouth from another process, it's
inconvenient to have to keep spawning the plymouth client binary
and keeping track of it - not to mention slow.
It's far cleaner to be able to link to the same boot client code
that the plymouth binary does, and communicate directly.
Place that code in a new libply-boot-client library, and install
the headers along with it.
Ray Strode [Thu, 25 Feb 2010 20:37:43 +0000 (15:37 -0500)]
[keyboard] Handle tty hangup better
We currently reconnect the terminal object on tty disconnects,
but we don't rewatch the keyboard. A disconnect will invalidate
the fd watch, so we need to handle it to prevent crashes.
Ray Strode [Tue, 26 Jan 2010 06:13:33 +0000 (01:13 -0500)]
[main] Defer password requests to client if daemon unavailable
There are times when plymouthd isn't in a position to ask for
the password (for instance, if the initramfs is about to run
init=/bin/bash or something). In those cases, any password
requests need to be handled by the client.
Ray Strode [Mon, 25 Jan 2010 22:07:56 +0000 (17:07 -0500)]
[client] Run ask-for-password command unconditionally
Even if we can't contact the daemon, we should still run the
ask-for-password command. This is because the command may
do things important for boot up to continue like unlocking
the root partition.
Charlie Brej [Mon, 25 Jan 2010 20:36:35 +0000 (20:36 +0000)]
[script] Use floor when implementing Int
Previously the Math.Int function converted to int and back to double.
The floor function works larger than those represented by integers and it
correctly handles NaN and Inf.
Ray Strode [Thu, 14 Jan 2010 22:05:15 +0000 (17:05 -0500)]
[branch-merge] Drop libpng requirements from base install
Right now, plymouthd links against libplybootsplash which links
against libpng. Only graphical splashes use libpng, but it's
always pulled in.
This merge splits libplybootsplash into two libraries: the core
bits, and the extractable graphical bits. Only the graphical splashes
link against the latter library. This way server installations can
get plymouth without pulling in libpng. This will reduce the amount
of security errata deployed to those server installations.
Ray Strode [Wed, 13 Jan 2010 19:27:46 +0000 (14:27 -0500)]
[splash] Rename libplybootsplash to libply-splash-core
Splashes aren't just shown for "boot" so including "boot" in
the name isn't right. Also, right now we link against libpng
even if we don't have any splash that uses it. This is the first
step toward making libpng only get linked to graphical splashes.
We'll need to move the graphical bits to their own library to
complete the process.
Ray Strode [Wed, 16 Dec 2009 18:20:34 +0000 (13:20 -0500)]
[branch-merge] Add man page
This merges the "man-page" branch to master.
Thanks to Adrian Glaubitz we now have a man page for plymouth.
This is a good start to documenting the basics of plymouth.
We'll eventually need more comprehensive reference documentation
before 1.0 is released, and now we have some of the ground work
(a docs/ directory) for that.