diff options
Diffstat (limited to 'driver/xscreensaver.c')
| -rw-r--r-- | driver/xscreensaver.c | 289 |
1 files changed, 196 insertions, 93 deletions
diff --git a/driver/xscreensaver.c b/driver/xscreensaver.c index b9c54f1..ce39c87 100644 --- a/driver/xscreensaver.c +++ b/driver/xscreensaver.c @@ -10,81 +10,155 @@ */ /* ======================================================================== + * HOW IT WORKS + * ======================================================================== + * * First we wait until the keyboard and mouse become idle for the specified - * amount of time. We do this in one of three different ways: periodically - * checking with the XIdle server extension; selecting key and mouse events - * on (nearly) all windows; or by waiting for the MIT-SCREEN-SAVER extension - * to send us a "you are idle" event. + * amount of time. Then, we map one or more full screen black windows. * - * Then, we map a full screen black window. + * We hold grabs on our override-redirect windows to ensure that user input + * does not leak through the screen saver to other applications on the + * desktop. If we cannot secure those grabs, we do not blank the screen. * - * We place a __SWM_VROOT property on this window, so that newly-started - * clients will think that this window is a "virtual root" window (as per - * the logic in the historical "vroot.h" header.) + * Creating those windows: * - * If there is an existing "virtual root" window (one that already had - * an __SWM_VROOT property) then we remove that property from that window. - * Otherwise, clients would see that window (the real virtual root) instead - * of ours (the impostor.) + * - If there are multiple screens, in the X11 sense of "screen", we + * create a window for each (the second number in the $DISPLAY + * variable, "host:display.screen".) * - * Then we pick a random program to run, and start it. Two assumptions - * are made about this program: that it has been specified with whatever - * command-line options are necessary to make it run on the root window; - * and that it has been compiled with vroot.h, so that it is able to find - * the root window when a virtual-root window manager (or this program) is - * running. + * - On Xinerama/RANDR systems, we create windows covering each physical + * monitor, meaning multiple windows on the same virtual desktop. * - * Then, we wait for keyboard or mouse events to be generated on the window. - * When they are, we kill the inferior process, unmap the window, and restore - * the __SWM_VROOT property to the real virtual root window if there was one. + * Then we pick a random program to run, and start it. We assume that it + * has been specified with whatever command line arguments are required to + * make it render onto the window we have provided. This is most easily + * accomplished by simply including the late-2003 version of "vroot.h", + * which will find the window via the $XSCREENSAVER_WINDOW environment + * variable, or via the __SWM_VROOT property. * - * On multi-screen systems, we do the above on each screen, and start - * multiple programs, each with a different value of $DISPLAY. + * Programs must draw onto the window provided. Creating their own window + * atop the xscreensaver window does not work. * - * On Xinerama systems, we do a similar thing, but instead create multiple - * windows on the (only) display, and tell the subprocess which one to use - * via the $XSCREENSAVER_WINDOW environment variable -- this trick requires - * a recent (Aug 2003) revision of vroot.h. + * Then, we wait for keyboard or mouse events to be generated on our + * windows. When they are, we kill the inferior process, unmap the + * windows, and restore the __SWM_VROOT property to the real virtual root + * window if there was one. * - * (See comments in screens.c for more details about Xinerama/RANDR stuff.) + * Do not ever kill XScreenSaver with signal 9 -- that can lead to the + * __SWM_VROOT property not being restored, and that will screw up your + * window manager. Rather than "kill", use "xscreensaver-command -exit" + * instead. * * While we are waiting for user activity, we also set up timers so that, * after a certain amount of time has passed, we can start a different * screenhack. We do this by killing the running child process with * SIGTERM, and then starting a new one in the same way. * - * If there was a real virtual root, meaning that we removed the __SWM_VROOT - * property from it, meaning we must (absolutely must) restore it before we - * exit, then we set up signal handlers for most signals (SIGINT, SIGTERM, - * etc.) that do this. Most Xlib and Xt routines are not reentrant, so it - * is not generally safe to call them from signal handlers; however, this - * program spends most of its time waiting, so the window of opportunity - * when code could be called reentrantly is fairly small; and also, the worst - * that could happen is that the call would fail. If we've gotten one of - * these signals, then we're on our way out anyway. If we didn't restore the - * __SWM_VROOT property, that would be very bad, so it's worth a shot. Note - * that this means that, if you're using a virtual-root window manager, you - * can really fuck up the world by killing this process with "kill -9". - * * This program accepts ClientMessages of type SCREENSAVER; these messages - * may contain the atoms ACTIVATE, DEACTIVATE, etc, meaning to turn the - * screensaver on or off now, regardless of the idleness of the user, - * and a few other things. The included "xscreensaver-command" program - * sends these messsages. + * may contain the atoms ACTIVATE, DEACTIVATE, etc. The included + * "xscreensaver-command" program sends these messsages. + * + * + * ======================================================================== + * ABOUT SERVER EXTENSIONS + * ======================================================================== + * + * Over the decades, there have been three X11 server extensions that are + * applicable to screen saving: + * + * - XIdle + * + * This extension provided a function to poll the user's idle time. + * It was simple and direct and worked great. Therefore, it was + * removed from the X11 distribution in 1994, with the release of + * X11R6. https://bugs.freedesktop.org/show_bug.cgi?id=1419 + * + * - SGI SCREEN_SAVER + * + * This extension sent two new events: "user is idle", and "user + * is no longer idle". It was simple and direct and worked great. + * But as the name implies, it only ever worked on Silicon + * Graphics machines. SGI became irrelevant around 1998 and went + * out of business in 2009. + * + * - MIT-SCREEN-SAVER + * + * This extension still exists, but it is useless to us. It takes + * the following approach: + * + * - When the user is idle, immediately map full screen black + * windows on each screen. + * + * - Inform the screen saver client that the screen is now black. + * + * - When user activity occurs, unmap the windows and then + * inform the screen saver client. + * + * The screen saver client can specify a few parameters of that + * window, like visual and depth, but that's it. + * + * The extension is designed with the assumption that a screen saver + * would render onto the provided window. However: + * + * - OpenGL programs may require different visuals than 2D X11 + * programs, and you can't change the visual of a window after it + * has been created. + * + * - The extension maps one window per X11 "Screen", which, in this + * modern world, tend to span the entire virtual desktop; whereas + * XScreenSaver runs savers full screen on each *monitor* instead. + * In other words, it is incompatible with Xinerama / RANDR. + * + * - Since this extension maps its own full-screen black windows and + * informs us of that after the fact, it prevents the "fade/unfade" + * animations from working properly. + * + * - Since it only tells us when the user is idle or non-idle, it + * prevents the "hysteresis" option from working properly (where + * we ignore tiny mouse motions). + * + * In summary, it creates its windows too early, removes them too late, + * creates windows of the wrong quantity and wrong shape, cannot create + * them with the proper visuals, and delivers too little information + * about what caused the user activity. + * + * Also my experience was that the MIT-SCREEN-SAVER extension was flaky, + * and using it at all led to frequent server crashes. * - * If we don't have the XIdle, MIT-SCREEN-SAVER, or SGI SCREEN_SAVER - * extensions, then we do the XAutoLock trick: notice every window that - * gets created, and wait 30 seconds or so until its creating process has - * settled down, and then select KeyPress events on those windows which - * already select for KeyPress events. It's important that we not select - * KeyPress on windows which don't select them, because that would - * interfere with event propagation. This will break if any program - * changes its event mask to contain KeyRelease or PointerMotion more than - * 30 seconds after creating the window, but such programs do not seem to - * occur in nature (I've never seen it happen in all these years.) + * So that's why, even if the server supports the MIT-SCREEN-SAVER + * extension, we don't use it. * - * The reason that we can't select KeyPresses on windows that don't have - * them already is that, when dispatching a KeyPress event, X finds the + * There is a recent trend among those who hack on video players to + * say, "We inhibit blanking by calling XResetScreenSaver and/or + * XScreenSaverSuspend, so our work here is done, fuck off." This + * betrays a willful ignorance of why the MIT-SCREEN-SAVER extension + * is a useless foundation upon which to build a screen saver or + * screen locker. + * + * The proper way to inhibit blanking while video is playing is to + * run "xscreensaver-command -deactivate" once a minute while the + * video is playing, as has been explained in the XScreenSaver FAQ + * for 3 decades: https://www.jwz.org/xscreensaver/faq.html#dvd + * + * The reason to do it as a heartbeat is so that your video player + * fails SAFE. Should the player exit abnormally, or freeze, the + * heartbeat stops coming, and screen blanking and locking can resume. + * + * + * ======================================================================== + * WITHOUT SERVER EXTENSIONS + * ======================================================================== + * + * So. Having established that there no longer exist any server extensions + * that help us in any way, here's how idle detection ACTUALLY works: + * + * XScreenSaver notices every window that gets created, wait 30 seconds or + * so until its birth pangs have settled down, and then select KeyPress + * events on those windows. + * + * Actually it only selects KeyPress events on windows that already selected + * for their own KeyPress events, because to do otherwise would interfere + * with event propagation. When dispatching a KeyPress event, X finds the * lowest (leafmost) window in the hierarchy on which *any* client selects * for KeyPress, and sends the event to that window. This means that if a * client had a window with subwindows, and expected to receive KeyPress @@ -94,7 +168,7 @@ * another client malfunction in this way. * * But here's a new kink that started showing up in late 2014: GNOME programs - * don't actually select for or receive KeyPress events! They do it behind + * don't actually select for or receive KeyPress events! They do it behind * the scenes through some kind of Input Method magic, even when running in * an en_US locale. However, in that case, those applications *do* seem to * update the _NET_WM_USER_TIME on their own windows every time they have @@ -103,44 +177,44 @@ * * To detect mouse motion, we periodically wake up and poll the mouse * position and button/modifier state, and notice when something has - * changed. We make this check every five seconds by default, and since the - * screensaver timeout has a granularity of one minute, this makes the + * changed. We make this check every five seconds by default, and since + * the screensaver timeout has a granularity of one minute, this makes the * chance of a false positive very small. We could detect mouse motion in * the same way as keyboard activity, but that would suffer from the same * "client changing event mask" problem that the KeyPress events hack does. - * I think polling is more reliable. + * Polling is more reliable. + * + * Also, because cats, trucks and earthquakes are things that exist, we + * ignore any mouse motions smaller than 10px by default. * - * On systems with /proc/interrupts (Linux) we poll that file and note when + * On Linux systems with /proc/interrupts we poll that file and note when * the interrupt counter numbers on the "keyboard" and "PS/2" lines change. - * (There is no reliable way, using /proc/interrupts, to detect non-PS/2 - * mice, so it doesn't help for serial or USB mice.) + * That was helpful for a decade or so, but then PS/2 devices went out of + * fashion, and there is no reliable way, using /proc/interrupts, to detect + * activity on USB keyboards or mice. Oh well. + * * - * None of this crap happens if we're using one of the extensions. Sadly, - * the XIdle extension hasn't been available for many years; the SGI - * extension only exists on SGIs; and the MIT extension, while widely - * deployed, is garbage in several ways. + * ======================================================================== + * DEBUGGING HINTS + * ======================================================================== * - * A third idle-detection option could be implemented (but is not): when - * running on the console display ($DISPLAY is `localhost`:0) and we're on a - * machine where /dev/tty and /dev/mouse have reasonable last-modification - * times, we could just stat() those. But the incremental benefit of - * implementing this is really small, so forget I said anything. + * - Have a second terminal handy, e.g. a Ctrl-Alt-F1 console. If you + * are running xscreensaver under a debugger, it might be helpful to + * run it from that other console, not from a terminal window within + * the same X11 session. That runs the risk of stopping at a + * breakpoint while the screen is blanked and the keyboard is grabbed. * - * Debugging hints: - * - Have a second terminal handy. - * - Be careful where you set your breakpoints, you don't want this to - * stop under the debugger with the keyboard grabbed or the blackout - * window exposed. - * - If you run your debugger under XEmacs, try M-ESC (x-grab-keyboard) - * to keep your emacs window alive even when xscreensaver has grabbed. * - Go read the code related to `debug_p'. + * * - You probably can't set breakpoints in functions that are called on * the other side of a call to fork() -- if your subprocesses are - * dying with signal 5, Trace/BPT Trap, you're losing in this way. - * - If you aren't using a server extension, don't leave this stopped - * under the debugger for very long, or the X input buffer will get - * huge because of the keypress events it's selecting for. This can - * make your X server wedge with "no more input buffers." + * dying with signal 5 Trace/BPT Trap, you're losing in this way. + * + * - Don't leave this stopped under the debugger for very long, or the + * X input buffer will get huge with un-read KeyPress events. This + * can make your X server hang with "no more input buffers." + * + * - Writing a new screen saver? See the README.hacking file. * * ======================================================================== */ @@ -418,10 +492,10 @@ saver_ehandler (Display *dpy, XErrorEvent *error) "#######################################################################\n" "\n" " If at all possible, please re-run xscreensaver with the command\n" - " line arguments `-sync -verbose -log log.txt', and reproduce this\n" - " bug. That will cause xscreensaver to dump a `core' file to the\n" - " current directory. Please include the stack trace from that core\n" - " file in your bug report. *DO NOT* mail the core file itself! That\n" + " line arguments \"-sync -log log.txt\", and reproduce this bug.\n" + " That will cause xscreensaver to dump a `core' file to the current\n" + " directory. Please include the stack trace from that core file\n" + " in your bug report. *DO NOT* mail the core file itself! That\n" " won't work. A \"log.txt\" file will also be written. Please *do*\n" " include the complete \"log.txt\" file with your bug report.\n" "\n" @@ -1314,6 +1388,7 @@ main_loop (saver_info *si) p->dpms_enabled_p && p->dpms_quickoff_p) { + /* Sync again just in case */ sync_server_dpms_settings (si->dpy, True, p->dpms_quickoff_p, p->dpms_standby / 1000, @@ -1431,6 +1506,24 @@ main_loop (saver_info *si) si->cycle_id = 0; cycle_timer ((XtPointer) si, 0); } + + /* If we are blanking only, optionally power down monitor + right now: quickoff means power down monitor again as soon + as unlock dialog is dismissed without unlocking. */ + if (!ok_to_unblank && + p->mode == BLANK_ONLY && + p->dpms_enabled_p && + p->dpms_quickoff_p) + { + /* Sync again just in case */ + sync_server_dpms_settings (si->dpy, True, + p->dpms_quickoff_p, + p->dpms_standby / 1000, + p->dpms_suspend / 1000, + p->dpms_off / 1000, + False); + monitor_power_on (si, False); + } } #endif /* !NO_LOCKING */ @@ -1496,6 +1589,7 @@ main (int argc, char **argv) saver_info *si = &the_si; saver_preferences *p = &si->prefs; struct passwd *spasswd; + Bool inhibit_stderr_capture_p = False; int i; /* It turns out that if we do setlocale (LC_ALL, "") here, people @@ -1554,10 +1648,14 @@ main (int argc, char **argv) shell = connect_to_server (si, &argc, argv); process_command_line (si, &argc, argv); - stderr_log_file (si); + inhibit_stderr_capture_p = stderr_log_file (si); print_banner (si); load_init_file(si->dpy, p); /* must be before initialize_per_screen_info() */ + + if (inhibit_stderr_capture_p && !p->verbose_p) + p->verbose_p = 1; /* "-log" implies "-verbose -no-capture-stderr" */ + blurb_timestamp_p = p->timestamp_p; /* kludge */ initialize_per_screen_info (si, shell); /* also sets si->fading_possible_p */ @@ -1601,13 +1699,18 @@ main (int argc, char **argv) p->dpms_off / 1000, False); - initialize_stderr (si); + initialize_stderr (si, inhibit_stderr_capture_p); handle_signals (si); store_saver_status (si); /* for xscreensaver-command -status */ # ifdef HAVE_LIBSYSTEMD /* Launch it in the background */ - si->systemd_pid = fork_and_exec_1 (si, 0, "xscreensaver-systemd"); + { + const char *cmd = (p->verbose_p + ? "xscreensaver-systemd -verbose" + : "xscreensaver-systemd"); + si->systemd_pid = fork_and_exec_1 (si, 0, cmd); + } # endif make_splash_dialog (si); @@ -1691,7 +1794,7 @@ XGetAtomName_safe (Display *dpy, Atom atom) } -static void +void clientmessage_response (saver_info *si, Window w, Bool error, const char *stderr_msg, const char *protocol_msg) |