* tentative merge of onelab myplc

[myplc.git] / host.init

diff --git a/host.init b/host.init
index f207e00..a299e82 100755 (executable)
--- a/host.init
+++ b/host.init
@@ -6,7 +6,7 @@
 #
 # description: Manages all PLC services on this machine
 #
-# $Id: host.init,v 1.8 2006/07/06 17:43:52 mlhuang Exp $
+# $Id: host.init 1078 2007-11-15 13:38:27Z thierry $
 #
 
 PATH=/sbin:/bin:/usr/bin:/usr/sbin
@@ -128,6 +128,58 @@ mountstatus_plc ()
     done
 }
 
+# safestop : tries to stop normally; if that fails, kills processes that are still using /plc 
+# needs the lsof rpm in the root context (should be a dependency of the myplc spec)
+function check_command ()
+{
+    command=$1; shift
+    found=$(type -p $command)
+    if [ -z "$found" ] ; then
+        echo "$COMMAND : requires command $command, was not found - exiting"
+       exit 1
+    fi
+}
+
+
+### when process stil use /plc/root, we cannot umount it
+function kill_all ()
+{
+    [ -n "$DEBUG" ] && set -x
+    check_command lsof
+    
+    echo -n "Killing processes using $PLC_ROOT and $PLC_DATA: "
+    # initialize process list
+    former_process_list="unlikely"
+      
+    # we ignore unknown uids for now, since we run in the chroot jail anyway
+    # not too sure about that though, 
+    while true; do
+        # get the list of processes - collapse and remove empty lines
+       process_list=$(lsof -t +D $PLC_ROOT +D $PLC_DATA)
+       if [ -z "$process_list" ] ; then
+           # we are done, let's bail out
+           success "$PLC_ROOT clear" ; echo ; return
+       fi
+       if [ "$process_list" = "$former_process_list" ] ; then
+           # we are stuck, no progress since last time : exit on error
+           failure "$PLC_ROOT locked" ; echo ; return
+       fi
+       # record for next loop
+       former_process_list="$process_list"
+       # kill them
+       kill $process_list
+       sleep 2
+       # check there are dead
+       for pid in $process_list ; do
+           ps -o pid $pid &> /dev/null 
+           if [ "$?" = 0 ] ; then
+               [ -n "$DEBUG" ] && echo "$pid survived kill - forcing kill -9"
+               kill -9 $pid
+           fi
+       done
+    done
+}
+
 # Get command
 shift $(($OPTIND - 1))
 command=$1
@@ -142,6 +194,7 @@ case "$command" in
 
     restart)
        stop $*
+       ERRORS=0
        start $*
        ;;
 
@@ -150,11 +203,28 @@ case "$command" in
        ;;      
 
     mount|umount|mountstatus)
-        ${command}_plc $*
+        ${command}_plc
+       ;;
+
+   kill)
+        kill_all
+        ;;
+
+   safestop)
+        stop
+       ### Checking : we might need to run kill
+       mounted=$(mountstatus_plc)
+       if [ -n "$mounted" ] ; then
+           echo "Umount failed : killing remaining processes and trying again"
+           ERRORS=0
+           kill_all
+           ERRORS=0
+           stop
+       fi
        ;;
   
     *)
-       echo "Usage: $0 {start|stop|restart|reload|mount|umount|mountstatus}"
+       echo "Usage: $0 {start|stop|restart|reload|mount|umount|mountstatus|kill|safestop}"
        RETVAL=1
        ;;
 esac