Merge branch 'adam-monit'

--Improve some init scripts
--Split start-daemons.sh.in
--Add monit conf files for HAL, RTUd and PPSi

doc/wrs-developer-manual.in

@@ -1814,6 +1814,47 @@ gentle restarts before.
 Date set from file @t{/update/saved_date} is never correct, but is based on
 best effort principle.
 
+@c ==========================================================================
+@node Supervision of running processes
+@section Supervision of running processes
+
+During normal operation @t{monit} supervises several processes running on a wrs
+switch. Check is done every 10 seconds. As for now supervised processes are:
+@t{wrsw_rtud},
+@t{wrsw_hal},
+@t{ppsi},
+@t{lighttpd},
+@t{dropbear},
+@t{snmpd}.
+
+In case any of the supervised processes does not run anymore (because of a crash,
+exit etc.), @t{monit} restarts missing process. If 5 restarts of a process
+occur during 10 cycles (10*10 seconds), the entire switch is restarted.
+
+Since @t{monit} is started from the inittab, even if @t{monit} crashes for some
+reason it will be re-spawned by the @t{init}.
+
+@c ==========================================================================
+@node Disabling monit
+@subsection Disabling monit
+
+In some cases, especially during development it is convenient to disable
+@t{monit} to avoid annoying re-spawns of the processes and restarts of the entire
+switch.
+@t{monit} can be disabled with command:
+@example
+/etc/init.d/monit.sh stop
+@end example
+which will send STOP signal to @t{monit} or by adding @t{CONFIG_MONIT_DISABLE=y} to dot-config.
+
+To re-enable @t{monit} first make sure there is no @t{CONFIG_MONIT_DISABLE=y} in dot-config, then execute command:
+@example
+/etc/init.d/monit.sh start
+@end example
+
+NOTE: Even when @t{monit} is disabled there is a process @t{/usr/bin/monit} in
+a process list, but its state is "stopped" (T).
+
 @c ==========================================================================
 @node SDB and Hardware Information
 @section SDB and Hardware Information

doc/wrs-user-manual.in

@@ -554,7 +554,7 @@ value is changed by the web interface, proper action is taken.
 @item CONFIG_MONIT_DISABLE
 
 	Disable monitoring of running processes by monit. Monit by default
-        re spawns processes that have died. This option is useful mostly during
+        re-spawns processes that have died. This option is useful mostly during
         development.
 
 @end table

userspace/rootfs_override/etc/init.d/dropbear

@@ -40,14 +40,27 @@ start() {
 	fi
 	umask 077
 	start-stop-daemon -S -q -p /var/run/dropbear.pid --exec /usr/sbin/dropbear
-	start_counter
-	echo "OK"
+	ret=$?
+	if [ $ret -eq 0 ]; then
+		start_counter
+		echo "OK"
+	elif [ $ret -eq 1 ]; then
+		echo "Failed (already running?)"
+	else
+		echo "Failed"
+	fi
 }
+
 stop() {
 	echo -n "Stopping dropbear sshd: "
 	start-stop-daemon -K -q -p /var/run/dropbear.pid
-	echo "OK"
+	if [ $? -eq 0 ]; then
+		echo "OK"
+	else
+		echo "Failed"
+	fi
 }
+
 restart() {
 	stop
 	start

userspace/rootfs_override/etc/init.d/hald.sh

+#!/bin/sh
+
+dotconfig=/wr/etc/dot-config
+
+start() {
+    echo -n "Starting HAL daemon: "
+
+    if [ -f $dotconfig ]; then
+	. $dotconfig
+    else
+	echo "$0 unable to source dot-config ($dotconfig)!"
+    fi
+
+    WRS_LOG=$CONFIG_WRS_LOG_HAL
+
+    # if empty turn it to /dev/null
+    if [ -z $WRS_LOG ]; then
+	WRS_LOG="/dev/null";
+    fi
+
+    # if a pathname, use it
+    if echo "$WRS_LOG" | grep / > /dev/null; then
+	eval LOGPIPE=\" \> $WRS_LOG 2\>\&1 \";
+    else
+	# not a pathname: use verbatim
+	eval LOGPIPE=\" 2\>\&1 \| logger -t wr-switch -p $WRS_LOG\"
+    fi
+
+# be carefull with pidof, no running script should have the same name as process
+    if pidof wrsw_hal > /dev/null; then
+	# wrsw_hal already running
+	echo "Failed (already running?)"
+    else
+	eval /wr/bin/wrsw_hal $LOGPIPE \&
+	echo "OK"
+    fi
+}
+
+stop() {
+    echo -n "Stopping HAL "
+    start-stop-daemon -K -q --exec /wr/bin/wrsw_hal
+    if [ $? -eq 0 ]; then
+	echo "OK"
+    else
+	echo "Failed"
+    fi
+}
+
+restart() {
+    stop
+    start
+}
+
+case "$1" in
+  start)
+	start
+	;;
+  stop)
+	stop
+	;;
+  restart|reload)
+	restart
+	;;
+  *)
+	echo $"Usage: $0 {start|stop|restart}"
+	exit 1
+	;;
+esac

userspace/rootfs_override/etc/init.d/lighttpd.sh

@@ -21,23 +21,27 @@ start() {
 	# running
 	start-stop-daemon -q -p /var/run/lighttpd.pid -S \
 		--exec /usr/sbin/lighttpd -- -f /var/www/lighttpd.config
-	if [ $? -eq 0 ]; then
+	ret=$?
+	if [ $ret -eq 0 ]; then
 		start_counter
 		echo "OK"
-	else
+	elif [ $ret -eq 1 ]; then
 		echo "Failed (already running?)"
+	else
+		echo "Failed"
 	fi
 }
+
 stop() {
 	echo -n "Stopping lighttpd daemon: "
 	start-stop-daemon -K -q -p /var/run/lighttpd.pid
 	if [ $? -eq 0 ]; then
-		start_counter
 		echo "OK"
 	else
 		echo "Failed"
 	fi
 }
+
 restart() {
 	stop
 	start

userspace/rootfs_override/etc/init.d/ppsi.sh

+#!/bin/sh
+
+dotconfig=/wr/etc/dot-config
+
+start() {
+    echo -n "Starting PTP daemon: "
+
+    if [ -f $dotconfig ]; then
+	. $dotconfig
+    else
+	echo "$0 unable to source dot-config ($dotconfig)!"
+    fi
+
+    WRS_LOG=$CONFIG_WRS_LOG_PTP
+
+    # if empty turn it to /dev/null
+    if [ -z $WRS_LOG ]; then
+	WRS_LOG="/dev/null";
+    fi
+
+    # if a pathname, use it
+    if echo "$WRS_LOG" | grep / > /dev/null; then
+	eval LOGPIPE=\" \> $WRS_LOG 2\>\&1 \";
+    else
+	# not a pathname: use verbatim
+	eval LOGPIPE=\" 2\>\&1 \| logger -t wr-switch -p $WRS_LOG\"
+    fi
+
+# be carefull with pidof, no running script should have the same name as process
+    if pidof ppsi > /dev/null; then
+	# ppsi already running
+	echo "Failed (already running?)"
+    else
+	eval /wr/bin/ppsi $LOGPIPE \&
+	echo "OK"
+    fi
+}
+
+stop() {
+    echo -n "Stopping PTP: "
+    start-stop-daemon -K -q --exec /wr/bin/ppsi
+    if [ $? -eq 0 ]; then
+	echo "OK"
+    else
+	echo "Failed"
+    fi
+}
+
+restart() {
+    stop
+    start
+}
+
+case "$1" in
+  start)
+	start
+	;;
+  stop)
+	stop
+	;;
+  restart|reload)
+	restart
+	;;
+  *)
+	echo $"Usage: $0 {start|stop|restart}"
+	exit 1
+	;;
+esac

userspace/rootfs_override/etc/init.d/reboot.sh

+#!/bin/sh
+
+# simple wrapper for monit to announce reboot to console
+echo "Monit triggered reboot due to $1" > /dev/console
+/sbin/reboot

userspace/rootfs_override/etc/init.d/rtud.sh

+#!/bin/sh
+
+dotconfig=/wr/etc/dot-config
+
+start() {
+    echo -n "Starting RTUd daemon: "
+
+    if [ -f $dotconfig ]; then
+	. $dotconfig
+    else
+	echo "$0 unable to source dot-config ($dotconfig)!"
+    fi
+
+    WRS_LOG=$CONFIG_WRS_LOG_RTU
+
+    # if empty turn it to /dev/null
+    if [ -z $WRS_LOG ]; then
+	WRS_LOG="/dev/null";
+    fi
+
+    # if a pathname, use it
+    if echo "$WRS_LOG" | grep / > /dev/null; then
+	eval LOGPIPE=\" \> $WRS_LOG 2\>\&1 \";
+    else
+	# not a pathname: use verbatim
+	eval LOGPIPE=\" 2\>\&1 \| logger -t wr-switch -p $WRS_LOG\"
+    fi
+
+# be carefull with pidof, no running script should have the same name as process
+    if pidof rtud > /dev/null; then
+	# rtud already running
+	echo "Failed (already running?)"
+    else
+	eval /wr/bin/wrsw_rtud $LOGPIPE \&
+	# ensure we receive UDP PTP frames, since ppsi supports UDP too.
+	/wr/bin/rtu_stat add 01:00:5e:00:01:81 18 0 &
+	echo "OK"
+    fi
+}
+
+stop() {
+    echo -n "Stopping RTUd "
+    start-stop-daemon -K -q --exec /wr/bin/wrsw_rtud
+    if [ $? -eq 0 ]; then
+	echo "OK"
+    else
+	echo "Failed"
+    fi
+}
+
+restart() {
+    stop
+    start
+}
+
+case "$1" in
+  start)
+	start
+	;;
+  stop)
+	stop
+	;;
+  restart|reload)
+	restart
+	;;
+  *)
+	echo $"Usage: $0 {start|stop|restart}"
+	exit 1
+	;;
+esac

userspace/rootfs_override/etc/init.d/snmp → userspace/rootfs_override/etc/init.d/snmpd

@@ -16,18 +16,31 @@ start_counter() {
 }
 
 start() {
+    echo -n "Starting snmpd daemon: "
     # Log to syslog at daemon level. And log source address (-a)
-    snmpd -Lsd -p $SNMP_PID -a -c $SNMP_CONF
-    start_counter
+    start-stop-daemon -q -p $SNMP_PID -S \
+		--exec /usr/sbin/snmpd -- -Lsd -p $SNMP_PID -a -c $SNMP_CONF
+    ret=$?
+    if [ $ret -eq 0 ]; then
+	start_counter
+	echo "OK"
+    elif [ $ret -eq 1 ]; then
+	echo "Failed (already running?)"
+    else
+	echo "Failed"
+    fi
 }
 
 stop() {
-    if pidof snmpd > /dev/null; then
-	echo -n "Stopping snmpd: "
-	start-stop-daemon -K -q -p $SNMP_PID
+    echo -n "Stopping snmpd: "
+    start-stop-daemon -K -q -p $SNMP_PID
+    if [ $? -eq 0 ]; then
 	echo "OK"
+    else
+	echo "Failed"
     fi
 }
+
 restart() {
     stop
     start

userspace/rootfs_override/etc/init.d/syslogd

@@ -17,9 +17,17 @@ start() {
     # Check the configuration file: if there's no target, don't run the thing.
     if ! grep -q '^##.*remote-host' /etc/rsyslog.conf; then
 	echo -n "Starting rsyslog: "
-	/usr/sbin/rsyslogd
-	start_counter
-	echo "OK"
+	start-stop-daemon -q -p /var/run/rsyslogd.pid -S \
+		--exec /usr/sbin/rsyslogd
+	ret=$?
+	if [ $ret -eq 0 ]; then
+	    start_counter
+	    echo "OK"
+	elif [ $ret -eq 1 ]; then
+	    echo "Failed (already running?)"
+	else
+	    echo "Failed"
+	fi
     else
 	echo "Not starting rsyslog: no target host selected"
 	# avoid SNMPd complaints about lack of start counter file
@@ -28,12 +36,15 @@ start() {
 }
 
 stop() {
-    if pidof rsyslogd > /dev/null; then
-	echo -n "Stopping rsyslog: "
-	start-stop-daemon -K -q -p /var/log/rsyslog.pid
+    echo -n "Stopping rsyslog: "
+    start-stop-daemon -K -q -p /var/run/rsyslogd.pid
+    if [ $? -eq 0 ]; then
 	echo "OK"
+    else
+	echo "Failed"
     fi
 }
+
 restart() {
     stop
     start

userspace/rootfs_override/etc/init.d/wrs_auxclk

@@ -34,10 +34,10 @@ if [ ! -z "$CONFIG_WRSAUXCLK_PPSHIFT" ]; then
 fi
 
 # execute wrs_auxclk
-echo "Configuring external clock clk2"
+echo -n "Configuring external clock clk2: "
 /wr/bin/wrs_auxclk $p_freq $p_duty $p_cshift $p_sigdel $p_ppshift > /dev/null 2>&1
 if [ $? == 0 ]; then
-	echo "External clock clk2 OK"
+	echo "OK"
 else
-	echo "Failed to configure external clock clk2!!!"
+	echo "Failed"
 fi

userspace/rootfs_override/etc/monit.d/dropbear.conf

@@ -2,4 +2,4 @@
    start program = "/etc/init.d/dropbear start"
    stop program = "/etc/init.d/dropbear stop"
    if failed host localhost port 22 type tcp then restart
-   if 5 restarts within 10 cycles then exec "/sbin/reboot"
+   if 5 restarts within 10 cycles then exec "/etc/init.d/reboot.sh dropbear"

userspace/rootfs_override/etc/monit.d/hald.conf

+ check process wrsw_hal matching /wr/bin/wrsw_hal
+   start program = "/etc/init.d/hald.sh start"
+   stop program = "/etc/init.d/hald.sh stop"
+   if 5 restarts within 10 cycles then exec "/etc/init.d/reboot.sh hald"

userspace/rootfs_override/etc/monit.d/lighttpd.conf

@@ -2,4 +2,4 @@
    start program = "/etc/init.d/lighttpd.sh start"
    stop program = "/etc/init.d/lighttpd.sh stop"
    if failed host localhost port 80 type tcp then restart
-   if 5 restarts within 10 cycles then exec "/sbin/reboot"
+   if 5 restarts within 10 cycles then exec "/etc/init.d/reboot.sh lighttpd"

userspace/rootfs_override/etc/monit.d/ppsi.conf

+ check process ppsi matching /wr/bin/ppsi
+   start program = "/etc/init.d/ppsi.sh start"
+   stop program = "/etc/init.d/ppsi.sh stop"
+   if 5 restarts within 10 cycles then exec "/etc/init.d/reboot.sh ppsi"
+   depends on wrsw_hal

userspace/rootfs_override/etc/monit.d/rtud.conf

+ check process wrsw_rtud matching /wr/bin/wrsw_rtud
+   start program = "/etc/init.d/rtud.sh start"
+   stop program = "/etc/init.d/rtud.sh stop"
+   if 5 restarts within 10 cycles then exec "/etc/init.d/reboot.sh rtud"
+   depends on wrsw_hal

userspace/rootfs_override/etc/monit.d/snmp.conf

- check process snmpd with pidfile /var/run/snmpd.pid
-   start program = "/etc/init.d/snmp start"
-   stop program = "/etc/init.d/snmp stop"
-   if 5 restart within 10 cycles then exec "/sbin/reboot"

userspace/rootfs_override/etc/monit.d/snmpd.conf

+ check process snmpd with pidfile /var/run/snmpd.pid
+   start program = "/etc/init.d/snmpd start"
+   stop program = "/etc/init.d/snmpd stop"
+   if 5 restart within 10 cycles then exec "/etc/init.d/reboot.sh snmpd"

userspace/rootfs_override/etc/rcS/S61hald.sh

+../init.d/hald.sh
\ No newline at end of file

userspace/rootfs_override/etc/rcS/S61ppsi.sh

+../init.d/ppsi.sh
\ No newline at end of file