Fix stalled ntpd startup on certain systems

2 files changed, 105 insertions, 60 deletions

diff --git a/recipes-navigation/gpsd/gpsd/gpsd_ubx_fixed.sh b/recipes-navigation/gpsd/gpsd/gpsd_ubx_fixed.sh
index c975942..6c9c126 100755
--- a/recipes-navigation/gpsd/gpsd/gpsd_ubx_fixed.sh
+++ b/recipes-navigation/gpsd/gpsd/gpsd_ubx_fixed.sh
@@ -5,6 +5,10 @@
 # This is to know when to set the system
 # and hardware clock, and when 
 # it is safe to start ntp.
+# DEBUG=1 to set debug
+NAME=gpsd_ubx_fixed
+shopt -s expand_aliases
+alias LG=logger\ -t\ ${NAME}
 if [[ -r /etc/default/gpsd ]] ; then
   . /etc/default/gpsd
 else
@@ -20,48 +24,62 @@ function rm_gps_file
     rm -f "${GPS_FIXFILE}"
   fi
 }
+function kill_it
+{
+    cmd=$1
+    pid=$2
+    sig=$3
+    ((DEBUG)) && LG -p user.info "terminating $cmd(${pid}) with SIG${sig}"
+    kill -SIG${sig} ${pid} >/dev/null 2>&1
+    result=$?                                                                                                                                                                                      
+    ((DEBUG)) && LG -p user.info "kill -SIG${sig} ${pid} result is $result"  
+}        
+
 
 # Most exits are errors, so remove the GPS fix file.
 trap rm_gps_file EXIT
 
 if ! [[ -x /usr/bin/gpsmon ]] ; then
-  logger -p user.err "gpsd_ubx_3dfix.sh:  Please install gpsmon"
+  LG -p user.err "Please install gpsmon"
   exit 0
 fi
 
-ppid=$$
-# echo gpsd_ubx_fixed parent $$ >>/tmp/gpsdlog.${ppid}
-# Restrict future kill to our process group.  During boot,
-# this is the entire rc complex, but not any daemons.
-pgid=$(ps -o pgrp -p $ppid --no-heading)
 # Terminate gpsmon after 10 seconds if it is our child
+# gpsmon has issues with terminals and stalling on terminal
+# read of the console.
 (
   sleep 10
-  # echo in child pgid $pgid  >>"/tmp/gpsdlog.${ppid}"
-  # ps -o  pid,pgid,args -e >>"/tmp/log1.${ppid}"
-  # Kill a gpsmon -a in our process group.  Hopefully this is our
-  # parents gpsmon -a.
-  gpsmonpid=$(ps -o pid,pgid,args -e | egrep "[[:space:]]${pgid}[[:space:]]+gpsmon[[:space:]]-a$" | sed -r 's/^[[:space:]]*([0-9]*)[[:space:]]+.*/\1/')
-  # echo sleeper TERM found pid $gpsmonpid >>"/tmp/gpsdlog.${ppid}"
-  if ((${#gpsmonpid})) ; then
-    logger -p user.info "terminating gpsmon(${gpsmonpid}) with SIGTERM"
-    # echo "terminating gpsmon(${gpsmonpid}) with SIGTERM"
-    kill ${gpsmonpid}
-  else
+  # We use BASHPID because our parent, $$, may have already exited, and
+  # then ps will not work.  BASHPID is the current subshell.
+  ppid=$BASHPID
+  pgid=$(ps --no-heading -o pgid -p $ppid)
+  
+  if ((${#pgid} == 0)) ; then
+    LG -p user.err "Could not find a pgid for $ppid"
+    ps -fjp $ppid | LG -p user.error
+  fi
+
+  gpsmonpid=$(pgrep -lg $pgid | egrep '[[:space:]]gpsmon$' | sed 's/[[:space:]].*//')
+  ((DEBUG)) && LG -p user.info "$pgid is pgid $gpsmonpid is gpsmonpid"
+  # Find the gpsmonpid that is our grandchild
+  ((didkill=0))
+  
+  # for loop is in case gpsmonpid has children.
+  for p in ${gpsmonpid} ; do
+    if ((${#p})) ; then
+        kill_it gpsmon $p TERM
+        ((didkill=p))
+        break
+    fi
+  done
+  if ((didkill == 0)) ; then
     exit 0
   fi
+  # Do a sigkill to be sure.
   sleep 2
-  # echo "SIGKILL next" >>"/tmp/gpsdlog.${ppid}"
-  # ps -o  pid,pgid,comm -e >>"/tmp/log2.${ppid}"
-  gpsmonpid=$(ps -o pid,pgid,args -e | egrep "[[:space:]]${pgid}[[:space:]]+gpsmon[[:space:]]-a$" | sed -r 's/^[[:space:]]*([0-9]*)[[:space:]]+.*/\1/')
-  # echo sleeper KILL found pid $gpsmonpid  >>"/tmp/gpsdlog.${ppid}"
-  # Kill with SIGKILL, in case SIGTERM fails.
-  if ((${#gpsmonpid})) ; then
-    logger -p user.info "terminating gpsmon(${gpsmonpid}) with SIGKILL"
-    kill -9 ${gpsmonpid}
-  fi
+  kill_it gpsmon $didkill KILL
 ) &
-  
+
 fix=""
 status=""
 
@@ -93,14 +111,13 @@ stuff="$(gpsmon -a 2>&1 | {
   fi
 })"
 
-# echo Past GPSMON in gpsd_ubx_fixed pid $$
 if ((${#stuff} == 0)) ; then
-  logger -p user.err No data from GPS
+  LG -p user.err "No data from GPS"
   exit 1
 fi
 
 if [[ -t 1 ]] ; then
-  stty echo icanon
+  stty echo icanon 2>/dev/null
 fi
 
 OIFS=${IFS}
@@ -119,8 +136,8 @@ case $status in
     ((fixOK=0))
     ;;
   *)
-    logger -p user.err "Is GPSD running?"
-    logger -p user.err "FIX OK field should be single hex digit: gpsmon data: $stuff"
+    LG -p user.err "Is GPSD running?"
+    LG -p user.err "FIX OK field should be single hex digit: gpsmon data: $stuff"
     exit 1
     break
     ;;
@@ -130,7 +147,7 @@ esac
 
 
 if ((fixOK == 0)) ; then
-  logger -p user.err "Problem with the GPS fix.  The fix is an even value, \"$status\", and should be an odd value.  fixOK=${fixOK}"
+  LG -p user.err "Problem with the GPS fix.  The fix is an even value, \"$status\", and should be an odd value.  fixOK=${fixOK}"
   exit 1
   # Retry later.  How?
 fi
@@ -140,14 +157,14 @@ fi
 for x in $GPSFIX ; do
   # echo test $x with $fix
   if [[ $x == $fix ]] ; then
-    logger user.info "GPS has fix $fix found in list GPSFIX: $GPSFIX"
+    LG -p user.info "GPS has fix $fix found in list GPSFIX: $GPSFIX"
     echo $x >"${GPS_FIXFILE}"
     GPS_FIXFILE=""
     exit 0
   fi
 done
 
-logger -p user.info "GPS fix is bad: $fix and should be one of: $GPSFIX"
+LG -p user.info "GPS fix is bad: $fix and should be one of: $GPSFIX"
 # Start later
 exit 1
 
diff --git a/recipes-navigation/gpsd/gpsd/gpsd_ubx_settime.sh b/recipes-navigation/gpsd/gpsd/gpsd_ubx_settime.sh
index 5644d57..f74e7b0 100755
--- a/recipes-navigation/gpsd/gpsd/gpsd_ubx_settime.sh
+++ b/recipes-navigation/gpsd/gpsd/gpsd_ubx_settime.sh
@@ -1,36 +1,63 @@
 #!/bin/bash
 # PPS= 1484247472.29561104 clock= 1484247990.00000000 offset= 517.704388959
+# DEBUG=1 to set debug
+NAME=gpsd_ubx_settime
+shopt -s expand_aliases
+alias LG="logger -t $NAME"
+function kill_it
+{
+    cmd=$1
+    pid=$2
+    sig=$3
+    ((DEBUG)) && LG -p user.info "terminating $cmd(${pid}) with SIG${sig}"
+    kill -SIG${sig} ${pid}
+    result=$?
+    ((DEBUG)) && LG -p user.info "kill -SIG${sig} ${pid} result is $result"
+}        
 
 trap "stty echo icanon" exit
 
 OIFS=$IFS
 IFS=$'\n'
-if ! /usr/sbin/gpsd_ubx_fixed ; then
-   logger -s -p user.warn "GPS does not have a fix yet.  Try again later."
+if ! /usr/sbin/gpsd_ubx_fixed </dev/null ; then
+   LG -s -p user.warn "GPS does not have a fix yet.  Try again later."
    exit 1
 fi
-# This shell script is shaped by two things:
-#   Lousy performance of sed and grep when used with gpsmon
-#   Need to terminate 
-# Terminate gpsmon after 10 seconds if it is our child          
-(                                                               
-  sleep 10                                                      
-  ppid=$$                                                       
-  gpsmonpid=$(ps -o pid,ppid,comm -e | egrep "[[:space:]]${ppid}[[:space:]]+gpsmon$" | sed -r 's/^[[:space:]]*([0-9]*)[[:space:]]+.*/\1/')
-  if ((${#gpsmonpid})) ; then                                                                                                             
-    logger -p user.info "terminating gpscat(${gpsmonpid}) with SIGTERM"                                                                   
-    kill ${gpsmonpid}                                                                                                                     
-  else                                                                                                                                    
-    exit 0                                                                                                                                
-  fi                                                                                                                                      
-  sleep 2                                                                                                                                 
-  gpsmonpid=$(ps -o pid,ppid,comm -e | egrep "[[:space:]]${ppid}[[:space:]]+gpsmon$" | sed -r 's/^[[:space:]]*([0-9]*)[[:space:]]+.*/\1/')
-  if ((${#gpsmonpid})) ; then                                                                                                             
-    logger -p user.info "terminating gpscat(${gpsmonpid}) with SIGKILL"                                                                   
-    kill -9 ${gpsmonpid}                                                                                                                  
-  fi                                                                                                                                      
-) &                                                                                                                                       
-                    
+
+
+# Terminate gpsmon after 10 seconds if it is our child
+# gpsmon has issues with terminals and stalling on terminal
+# read of the console.
+(
+  sleep 10
+  ppid=$BASHPID
+  pgid=$(ps --no-heading -o pgid -p $ppid)
+  
+  if ((${#pgid} == 0)) ; then
+    LG -p user.err "Could not find a pgid for $ppid"
+    ps -fjp $ppid | LG -p user.error
+  fi
+
+  gpsmonpid=$(pgrep -lg $pgid | egrep '[[:space:]]gpsmon$' | sed 's/[[:space:]].*//')
+  ((DEBUG)) && LG -p user.info "$pgid is pgid $gpsmonpid is gpsmonpid"
+  # Find the gpsmonpid that is our grandchild
+  ((didkill=0))
+  
+  # for loop is in case gpsmonpid has children.
+  for p in ${gpsmonpid} ; do
+    if ((${#p})) ; then
+        kill_it gpsmon $p TERM
+        ((didkill=p))
+        break
+    fi
+  done
+  if ((didkill == 0)) ; then
+    exit 0
+  fi
+  # Do a sigkill to be sure.
+  sleep 2
+  kill_it gpsmon $didkill KILL
+) &
                     
 epoch=$(gpsmon -a 2>&1 | ( while read ln ; do
 	if [[ $ln =~ ^[[:space:]]*PPS=.*clock=[[:space:]]*([0-9]*)\. ]] ; then
@@ -39,9 +66,10 @@ epoch=$(gpsmon -a 2>&1 | ( while read ln ; do
 	fi
 done ))
 if [[ $epoch =~ ^[0-9]+$ ]] ; then
+  LG -p user.warn "Updating time to epoch time, ${epoch}."
   date +%s -s @${epoch} >/dev/null
 else
-  logger user.err "gpsmon output is bad."
+  LG -p user.err "gpsmon output is bad."
   exit 1
 fi
 exit 0