diff options
Diffstat (limited to 'kdesktop/xautolock_engine.c')
-rw-r--r-- | kdesktop/xautolock_engine.c | 419 |
1 files changed, 419 insertions, 0 deletions
diff --git a/kdesktop/xautolock_engine.c b/kdesktop/xautolock_engine.c new file mode 100644 index 000000000..1825b7fe2 --- /dev/null +++ b/kdesktop/xautolock_engine.c @@ -0,0 +1,419 @@ +/***************************************************************************** + * + * Authors: Michel Eyckmans (MCE) & Stefan De Troch (SDT) + * + * Content: This file is part of version 2.x of xautolock. It implements + * the program's core functions. + * + * Please send bug reports etc. to eyckmans@imec.be. + * + * -------------------------------------------------------------------------- + * + * Copyright 1990,1992-1999,2001-2002 by Stefan De Troch and Michel Eyckmans. + * + * Versions 2.0 and above of xautolock are available under version 2 of the + * GNU GPL. Earlier versions are available under other conditions. For more + * information, see the License file. + * + *****************************************************************************/ + +#include <X11/Xlib.h> +#include <time.h> + +#include "xautolock_c.h" + +/* + * Function for querying the idle time from the server. + * Only used if either the Xidle or the Xscreensaver + * extension is present. + */ +void +xautolock_queryIdleTime (Display* d) +{ + Time idleTime = 0; /* millisecs since last input event */ + +#ifdef HasXidle + if (xautolock_useXidle) + { + XGetIdleTime (d, &idleTime); + } + else +#endif /* HasXIdle */ + { +#ifdef HasScreenSaver + if( xautolock_useMit ) + { + static XScreenSaverInfo* mitInfo = 0; + if (!mitInfo) mitInfo = XScreenSaverAllocInfo (); + XScreenSaverQueryInfo (d, DefaultRootWindow (d), mitInfo); + idleTime = mitInfo->idle; + } + else +#endif /* HasScreenSaver */ + { + d = d; /* shut up */ + return; /* DIY */ + } + } + + if (idleTime < CHECK_INTERVAL ) + { + xautolock_resetTriggers (); + } +} + +/* + * Function for monitoring pointer movements. This implements the + * `corners' feature and as a side effect also tracks pointer + * related user activity. The latter actually is only needed when + * we're using the DIY mode of operations, but it's much simpler + * to do it unconditionally. + */ +void +xautolock_queryPointer (Display* d) +{ + Window dummyWin; /* as it says */ + int dummyInt; /* as it says */ + unsigned mask; /* modifier mask */ + int rootX; /* as it says */ + int rootY; /* as it says */ + int corner; /* corner index */ + time_t now; /* as it says */ + time_t newTrigger; /* temporary storage */ + int i; /* loop counter */ + static Window root; /* root window the pointer is on */ + static Screen* screen; /* screen the pointer is on */ + static unsigned prevMask = 0; /* as it says */ + static int prevRootX = -1; /* as it says */ + static int prevRootY = -1; /* as it says */ + static Bool firstCall = True; /* as it says */ + + /* + * Have a guess... + */ + if (firstCall) + { + firstCall = False; + root = DefaultRootWindow (d); + screen = ScreenOfDisplay (d, DefaultScreen (d)); + } + + /* + * Find out whether the pointer has moved. Using XQueryPointer for this + * is gross, but it also is the only way never to mess up propagation + * of pointer events. + */ + if (!XQueryPointer (d, root, &root, &dummyWin, &rootX, &rootY, + &dummyInt, &dummyInt, &mask)) + { + /* + * Pointer has moved to another screen, so let's find out which one. + */ + for (i = -1; ++i < ScreenCount (d); ) + { + if (root == RootWindow (d, i)) + { + screen = ScreenOfDisplay (d, i); + break; + } + } + } + + if ( rootX == prevRootX + && rootY == prevRootY + && mask == prevMask) + { + xautolock_corner_t* corners = xautolock_corners; + /* + * If the pointer has not moved since the previous call and + * is inside one of the 4 corners, we act according to the + * contents of the "corners" array. + * + * If rootX and rootY are less than zero, don't lock even if + * ca_forceLock is set in the upper-left corner. Why? 'cause + * on initial server startup, if (and only if) the pointer is + * never moved, XQueryPointer() can return values less than + * zero (only some servers, Openwindows 2.0 and 3.0 in + * particular). + */ + if ( (corner = 0, + rootX <= cornerSize && rootX >= 0 + && rootY <= cornerSize && rootY >= 0) + || (corner++, + rootX >= WidthOfScreen (screen) - cornerSize - 1 + && rootY <= cornerSize) + || (corner++, + rootX <= cornerSize + && rootY >= HeightOfScreen (screen) - cornerSize - 1) + || (corner++, + rootX >= WidthOfScreen (screen) - cornerSize - 1 + && rootY >= HeightOfScreen (screen) - cornerSize - 1)) + { + now = time (0); + + switch (corners[corner]) + { + case ca_forceLock: +#if 0 + newTrigger = now + (useRedelay ? cornerRedelay : cornerDelay) - 1; +#else + newTrigger = now; +#endif + +#if 0 + if (newTrigger < lockTrigger) + { + setLockTrigger (newTrigger - now); + } +#else + xautolock_setTrigger( newTrigger ); +#endif + break; + + case ca_dontLock: + xautolock_resetTriggers (); + +#ifdef __GNUC__ + default: ; /* Makes gcc -Wall shut up. */ +#endif /* __GNUC__ */ + } + } + } + else + { +#if 0 + useRedelay = False; +#endif + prevRootX = rootX; + prevRootY = rootY; + prevMask = mask; + + xautolock_resetTriggers (); + } +} + +#if 0 +/* + * Support for deciding whether to lock or kill. + */ +void +evaluateTriggers (Display* d) +{ + static time_t prevNotification = 0; + time_t now = 0; + + /* + * Obvious things first. + * + * The triggers are being moved all the time while in disabled + * mode in order to make absolutely sure we cannot run into + * trouble by an enable message coming in at an odd moment. + * Otherwise we possibly might lock or kill too soon. + */ + if (disabled) + { + resetTriggers (); + } + + /* + * Next, wait for (or kill, if we were so told) the previous + * locker (if any). Note that this must also be done while in + * disabled mode. Not only to avoid a potential zombie proces + * hanging around until we are re-enabled, but also to prevent + * us from incorrectly setting a kill trigger at the moment + * when we are finally re-enabled. + */ +#ifdef VMS + if (vmsStatus == 0) + { +#else /* VMS */ + if (lockerPid) + { +#if !defined (UTEKV) && !defined (SYSV) && !defined (SVR4) + union wait status; /* childs process status */ +#else /* !UTEKV && !SYSV && !SVR4 */ + int status = 0; /* childs process status */ +#endif /* !UTEKV && !SYSV && !SVR4 */ + + if (unlockNow && !disabled) + { + (void) kill (lockerPid, SIGTERM); + } + +#if !defined (UTEKV) && !defined (SYSV) && !defined (SVR4) + if (wait3 (&status, WNOHANG, 0)) +#else /* !UTEKV && !SYSV && !SVR4 */ + if (waitpid (-1, &status, WNOHANG)) +#endif /* !UTEKV && !SYSV && !SVR4 */ + { + /* + * If the locker exited normally, we disable any pending kill + * trigger. Otherwise, we assume that it either has crashed or + * was not able to lock the display because of an existing + * locker (which may have been started manually). In both of + * the later cases, disabling the kill trigger would open a + * loop hole. + */ + if ( WIFEXITED (status) + && WEXITSTATUS (status) == EXIT_SUCCESS) + { + disableKillTrigger (); + } + + useRedelay = True; + lockerPid = 0; + } +#endif /* VMS */ + + setLockTrigger (lockTime); + + /* + * No return here! The pointer may be sitting in a corner, while + * parameter settings may be such that we need to start another + * locker without further delay. If you think this cannot happen, + * consider the case in which the locker simply crashed. + */ + } + + unlockNow = False; + + /* + * Note that the above lot needs to be done even when we're in + * disabled mode, since we may have entered said mode with an + * active locker around. + */ + if (disabled) return; + + /* + * Is it time to run the killer command? + */ + now = time (0); + + if (killTrigger && now >= killTrigger) + { + /* + * There is a dirty trick here. On the one hand, we don't want + * to block until the killer returns, but on the other one + * we don't want to have it interfere with the wait() stuff we + * do to keep track of the locker. To obtain both, the killer + * command has already been patched by KillerChecker() so that + * it gets backgrounded by the shell started by system(). + * + * For the time being, VMS users are out of luck: their xautolock + * will indeed block until the killer returns. + */ + (void) system (killer); + setKillTrigger (killTime); + } + + /* + * Now trigger the notifier if required. + */ + if ( notifyLock + && now + notifyMargin >= lockTrigger + && prevNotification < now - notifyMargin - 1) + { + if (notifierSpecified) + { + /* + * Here we use the same dirty trick as for the killer command. + */ + (void) system (notifier); + } + else + { + (void) XBell (d, bellPercent); + (void) XSync (d, 0); + } + + prevNotification = now; + } + + /* + * Finally fire up the locker if time has somehow come. + */ + if ( lockNow + || now >= lockTrigger) + { +#ifdef VMS + if (vmsStatus != 0) +#else /* VMS */ + if (!lockerPid) +#endif /* VMS */ + { + switch (lockerPid = vfork ()) + { + case -1: + lockerPid = 0; + break; + + case 0: + (void) close (ConnectionNumber (d)); +#ifdef VMS + vmsStatus = 0; + lockerPid = lib$spawn ((lockNow ? &nowLockerDescr : &lockerDescr), + 0, 0, &1, 0, 0, &vmsStatus); + + if (!(lockerPid & 1)) exit (lockerPid); + +#ifdef SLOW_VMS + (void) sleep (SLOW_VMS_DELAY); +#endif /* SLOW_VMS */ +#else /* VMS */ + (void) execl ("/bin/sh", "/bin/sh", "-c", + (lockNow ? nowLocker : locker), 0); +#endif /* VMS */ + _exit (EXIT_FAILURE); + + default: + /* + * In general xautolock should keep its fingers off the real + * screensaver because no universally acceptable policy can + * be defined. In no case should it decide to disable or enable + * it all by itself. Setting the screensaver policy is something + * the locker should take care of. After all, xautolock is not + * supposed to know what the "locker" does and doesn't do. + * People might be using xautolock for totally different + * purposes (which, by the way, is why it will accept a + * different set of X resources after being renamed). + * + * Nevertheless, simply resetting the screensaver is a + * convenience action that aids many xlock users, and doesn't + * harm anyone (*). The problem with older versions of xlock + * is that they can be told to replace (= disable) the real + * screensaver, but forget to reset that same screensaver if + * it was already active at the time xlock starts. I guess + * xlock initially wasn't designed to be run without a user + * actually typing the comand ;-). + * + * (*) Well, at least it used not to harm anyone, but with the + * advent of DPMS monitors, it now can mess up the power + * saving setup. Hence we better make it optional. + * + * Also, some xlock versions also unconditionally call + * XResetScreenSaver, yielding the same kind of problems + * with DPMS that xautolock did. The latest and greatest + * xlocks also have a -resetsaver option for this very + * reason. You may want to upgrade. + */ + if (resetSaver) (void) XResetScreenSaver(d); + + setLockTrigger (lockTime); + (void) XSync (d,0); + } + + /* + * Once the locker is running, all that needs to be done is to + * set the killTrigger if needed. Notice that this must be done + * even if we actually failed to start the locker. Otherwise + * the error would "propagate" from one feature to another. + */ + if (killerSpecified) setKillTrigger (killTime); + + useRedelay = False; + } + + lockNow = False; + } +} +#endif |