Fix chat_wrapper to handle COPS=2, CFUN=0 state

1 files changed, 158 insertions, 24 deletions

diff --git a/recipes-core/multitech/config/chat_wrapper b/recipes-core/multitech/config/chat_wrapper
index f1fd30c..b0bf9e5 100755
--- a/recipes-core/multitech/config/chat_wrapper
+++ b/recipes-core/multitech/config/chat_wrapper
@@ -18,30 +18,117 @@
 # If there are not #MT AT+CGDCONT= lines, then any line without
 # a comment chararacter before AT+CGDONT= is accepted, but only the
 # last one in the file.
+#
+# On entry to this script, file descriptor 0 is the modem.
+# For ppp to work, stty must work on file descriptor 0.
+# If stty fails, so will ppp.
 
 NAME=chat_wrapper
 CONFIG=/etc/default/${NAME}
+CHAT_WRAPPER_RECURSION=/run/chat_wrapper_recursion
+
 function finish
 {
         ${LOG} "Launch:" "$@"
-	exec "$@"
+        exec "$@"
         # NOTREACHED
 }
 
+# This function is used to determine if the modem and serial driver
+# are working.  In testing, the driver has gotten into states where
+# stty fails.  Before the chat script is started, the ioctl TCGETS is
+# performed, and will fail if the driver is hung.  This same function
+# is called by stty.  The modem port used by pppd is STDIN.  Cases
+# have been observed on the H5 where radio-cmd works, but stty does
+# not, resulting in pppd failing. When this happens, radio-reset is
+# the best thing to try.
+function sanity
+{
+  if ! log=$(stty 2>&1) || ! [[ $(radio-cmd -t10000 'AT') =~ ^OK ]]; then
+    ${LOG} "Modem has issues, so see if we can try again"
+    ${LOG} "stty error: $(log)"
+    array=( )
+    # our parent is PPID, which is pppd.  Get the pppd
+    # parameter list.
+    while IFS= read -r -d '' eparm; do
+        execarg+=( "$eparm" )
+    done < <(cat /proc/$PPID/cmdline)
+    i=0
+    ${LOG} "List parameters"
+    while ((i < ${#execarg[@]})) ; do
+       ${LOG} "$i: ${execarg[$i]}"
+       ((i++))
+    done
+    if [[ ${execarg[0]} == "pppd" ]] ; then
+      execarg[0]=/usr/sbin/pppd
+    fi
+
+    # Break any recursion loop.
+    if [[ -f ${CHAT_WRAPPER_RECURSION} ]] ; then
+       ${LOG}  "Failure: modem is still unresponsive after reset"
+       rm -f ${CHAT_WRAPPER_RECURSION}
+       exit 1
+    fi
+    touch ${CHAT_WRAPPER_RECURSION}
+    ${LOG} "Bad TTY, possibly due to modem in bad state"
+    ${LOG} "Reset modem, and we relaunch pppd"
+    # Our terminal is the modem.  When we reset the modem,
+    # we will be sent a SIGHUP.  Ignore it (:).
+    trap : SIGHUP
+    /usr/sbin/mts-io-sysfs store radio-reset 0
+    sleep 5
+    count=0
+    while ((count < 60)) ; do
+      # Wait for radio-reset to get the modem responding
+      if [[ $(radio-cmd -t10000 'AT') =~ ^OK ]]; then
+        # parent is dying, so remove the pppd locks for the new pppd.
+        for f in /run/lock/LCK..modem_at0 /run/lock/LCK..modem_at1 ; do
+          if [[ -f $f ]] ; then
+            mypid=$(cat $f)
+            ${LOG} "PPID is $PPID, pid is $mypid"
+            # Remove the lock, if it belongs to our parent, pppd
+            if [ $mypid = $PPID ] ; then
+              ${LOG} "match $mypid $PPID"
+              sleep 60
+              rm -f $f
+            else
+              ${LOG} "mismatch |$mypid| |$PPID|"
+            fi
+          fi
+        done
+        # Locks gone, so start new pppd
+        nohup "${execarg[@]}" >/dev/null 2>&1 &
+        # Tell our parent we failed, which causes our pppd to terminate.
+        exit 1
+      else
+        sleep 5
+      fi
+    done
+    ${LOG} "ERROR: After five minutes the modem is still not ready"
+    exit 1
+  fi
+  return 0
+} # End of sanity function
+
+# If > 0, need to wait because we were not registered.
+WAITREG=0
+
 [[ -f $CONFIG ]] || exit 1
 
 . ${CONFIG} 
 
 : ${REGWAITTIME:=300}
 : ${FINALWAIT:=5}
+: ${COPSWAIT:=10}
+: ${CFUNWAIT:=10}
 
 : ${LOG:="/usr/bin/logger -t ${NAME} -p daemon.notice"}
 
 ${LOG} Timeout is $REGWAITTIME, execute "$@" 
 ((i=$#))
 chatscript="${!i}"
-${LOG} Parsing chat script "$chatscript"
 
+${LOG} Parsing chat script "$chatscript"
 # CONTEXT is last context string in chat script
 CONTEXT=$(egrep "^#MT[[:space:]]+(AT\+CGDCONT=.*)" ${chatscript} | tail -1)
 if ((${#CONTEXT} == 0)) ; then
@@ -60,9 +147,25 @@ if ((${#CONTEXT} == 0)) ; then
 fi
 ((CONTEXTNUM=${BASH_REMATCH[2]}))
 
+# Make sure CFUN=1, in case the last pppd was aborted with it set to 0.
+${LOG} "Verify CFUN=1"
+cfun=$(/usr/bin/radio-cmd ${RADIOOPTION} -t10000 'AT+CFUN?')
+result1=$?
+if ((result1 != 0)) || ((${#cfun} == 0)) ; then
+   ${LOG} "Is modem in bad state -- \"CFUN?\" is broken?"
+   sanity
+   ((WAITREG++))
+fi
+if ! [[ $cfun =~ ^\+CFUN:\ 1[,[:space:]] ]] ; then
+    /usr/bin/radio-cmd ${RADIOOPTION} -t10000 'AT+CFUN=1'
+    ${LOG} "Just set CFUN=1 so wait ${CFUNWAIT} seconds"
+    sleep ${CFUNWAIT}
+    ((WAITREG++))
+fi
 
 ${LOG} "Using Context ${CONTEXTNUM} based on chat script: ${CONTEXT}"
 
+
 # At this point if there is no context number, we can skip everything else.
 
 # Get Modem's context settings
@@ -137,7 +240,15 @@ if [[ $MPDP != $PDP ]] || [[ $MAPN != $APN ]] || \
    ${LOG} "$MCONTEXT"
    ${LOG} "Dropping registration with carrier to set context"
    # Need to deregister
+   ((WAITREG++))
    /usr/bin/radio-cmd ${RADIOOPTION} -t10000 'AT+COPS=2'
+
+   count=10
+   while ((count > 0)) && ! [[ $(/usr/bin/radio-query ${RADIOOPTION} --netreg) =~ ^NOT\ REGISTERED$ ]] ; do
+      sleep 2
+      ((count--))  
+   done
+   
    # H5 radios do not like addresses that are empty with ""
    if ((${#ADDR} == 0)) ; then
      CMDSTR="AT+CGDCONT=${CONTEXTNUM},\"${PDP}\",\"${APN}\",,$DCOMP,${HCOMP}${FULLBOAT}"
@@ -154,7 +265,7 @@ if [[ $MPDP != $PDP ]] || [[ $MAPN != $APN ]] || \
      ((count--)) || true
      LOGMSG=$(/usr/bin/radio-cmd ${RADIOOPTION} -t10000 "${CMDSTR}" 2>&1)
      result=$?
-     ${LOG} "Result ${result}, Got response ${LOGMSG}"
+     ${LOG} "Setting context: Result ${result}, Got response ${LOGMSG}"
      if [[ $LOGMSG =~ ERROR ]] ; then
 	result=1
      fi
@@ -163,13 +274,16 @@ if [[ $MPDP != $PDP ]] || [[ $MAPN != $APN ]] || \
 
    # re-enable the modem to defaults.
    /usr/bin/radio-cmd ${RADIOOPTION} -t10000 'AT+COPS=0'
-   sleep 1
 
    # Some older modems will not re-register after the COPS
    # command if you do not reset them using CFUN.
    /usr/bin/radio-cmd ${RADIOOPTION} -t10000 'AT+CFUN=0'
+   ${LOG} "Set COPS=0, CFUN=1, so wait ${CFUNWAIT} seconds"
+   sleep ${CFUNWAIT}
    /usr/bin/radio-cmd ${RADIOOPTION} -t10000 'AT+CFUN=1'
 
+   sleep ${CFUNWAIT}
+
    # Abort PPP if the Context is not correct.  We don't want
    # to cause the carrier to disable the account by dialing
    # out with a bad APN.
@@ -179,27 +293,47 @@ if [[ $MPDP != $PDP ]] || [[ $MAPN != $APN ]] || \
    fi
    sleep 1
    ${LOG} "New context is set.  Wait up to $REGWAITTIME seconds to register"
-   # Wait for registration
-   uptime=$(cat /proc/uptime)
-   [[ $uptime =~ ^([^\.]*) ]]
-   t0=${BASH_REMATCH[1]}
-   while ! [[ $(/usr/bin/radio-query ${RADIOOPTION} --netreg) =~ ^REGISTERED$ ]] ; do
-      sleep 5
-      uptime=$(cat /proc/uptime)
-      [[ $uptime =~ ^([^\.]*) ]]
-      t1=${BASH_REMATCH[1]}
-      if ((t1-t0 > REGWAITTIME)) ; then
-	   ${LOG} "$((t1-t0)) seconds has expired"
-	   exit 1
-      fi	
-   done
-   uptime=$(cat /proc/uptime)
-   [[ $uptime =~ ^([^\.]*) ]]
-   t1=${BASH_REMATCH[1]}
-   ${LOG} "Re-registered in $((t1-t0)) seconds -- wait ${FINALWAIT} more seconds"
-   sleep $FINALWAIT
 else
-   ${LOG} "Context $CONTEXTNUM matches and nothing to do."
+    ${LOG} "Context $CONTEXTNUM matches  verify COPS, then wait for registration."
+    #  Check and set COPS and CFUN to be sure we were not interrupted at some point
+    # in the past.
+    cops=$(/usr/bin/radio-cmd ${RADIOOPTION} -t10000 'AT+COPS?')
+    result0=$?
+
+    if ((result0 != 0)) || ! [[ $cops =~ ^\+COPS:\ 0, ]] ; then
+        /usr/bin/radio-cmd ${RADIOOPTION} -t10000 'AT+COPS=0'
+        ((WAITREG++))
+        ${LOG} "Just set COPS=0 so wait ${COPSWAIT} seconds"
+        sleep ${COPSWAIT}
+    fi
+
+fi
+# Wait for registration
+uptime=$(cat /proc/uptime)
+[[ $uptime =~ ^([^\.]*) ]]
+t0=${BASH_REMATCH[1]}
+while ! [[ $(/usr/bin/radio-query ${RADIOOPTION} --netreg) =~ ^REGISTERED$ ]] ; do
+    sleep 5
+    ((WAITREG++))
+    uptime=$(cat /proc/uptime)
+    [[ $uptime =~ ^([^\.]*) ]]
+    t1=${BASH_REMATCH[1]}
+    if ((t1-t0 > REGWAITTIME)) ; then
+        ${LOG} "$((t1-t0)) seconds has expired without registration"
+        sanity
+    fi	
+done
+uptime=$(cat /proc/uptime)
+[[ $uptime =~ ^([^\.]*) ]]
+t1=${BASH_REMATCH[1]}
+${LOG} "Re-registered in $((t1-t0)) seconds"
+if ((WAITREG > 0)) ; then
+  ${LOG} "Wait ${FINALWAIT} more seconds"
+  sleep $FINALWAIT
 fi
+sanity
+
+rm -f ${CHAT_WRAPPER_RECURSION}
 finish "$@"
+
 # NOTREACHED