Allow zfsbootmenu to takeover a running instance

Instead of requiring a user to rm /zfsbootmenu/active, running 'zfsbootmenu' or 'zbm' from the recovery shell will automatically stop an already-running copy of /bin/zfsbootmenu. It will also block the auto-boot countdown sequence that might be executing on the main tty (tty0, ttyS0, etc). 

If the running instance can't be killed for some reason (/zfsbootmenu/active still exists), zfsbootmenu will spin with the option to exit to a recovery shell until that file is removed.

90zfsbootmenu/zfsbootmenu-init.sh

@@ -19,10 +19,6 @@ if ! is_lib_sourced > /dev/null 2>&1 ; then
   exec /bin/bash
 fi
 
-if [ -z "${BASE}" ]; then
-  export BASE="/zfsbootmenu"
-fi
-
 mkdir -p "${BASE}"
 
 # Attempt to load spl normally
@@ -189,11 +185,14 @@ fi
 : > "${BASE}/initialized"
 
 # If BOOTFS is not empty display the fast boot menu
-if [ -n "${BOOTFS}" ]; then
+# shellcheck disable=SC2154
+if [ "${menu_timeout}" -ge 0 ] && [ -n "${BOOTFS}" ]; then
   # Draw a countdown menu
   # shellcheck disable=SC2154
-  if [ "${menu_timeout}" -ge 0 ]; then
-    if delay="${menu_timeout}" prompt="Booting ${BOOTFS} in %0.${#menu_timeout}d seconds" timed_prompt "[ENTER] to boot" "[ESC] boot menu" ; then
+  if delay="${menu_timeout}" prompt="Booting ${BOOTFS} in %0.${#menu_timeout}d seconds" timed_prompt "[ENTER] to boot" "[ESC] boot menu" ; then
+    # This lock file is present if someone has SSH'd to take control
+    # Do not attempt to automatically boot if present
+    if [ ! -e "${BASE}/active" ] ; then
       # Clear screen before a possible password prompt
       tput clear
       if ! NO_CACHE=1 load_key "${BOOTFS}"; then
@@ -206,6 +205,12 @@ if [ -n "${BOOTFS}" ]; then
   fi
 fi
 
+# If the lock file is present, drop to a recovery shell to avoid
+# stealing control back from an SSH session
+if [ -e "${BASE}/active" ] ; then
+  emergency_shell "type 'exit' to return to ZFSBootMenu"
+fi
+
 while true; do
   if [ -x /bin/zfsbootmenu ]; then
     /bin/zfsbootmenu

90zfsbootmenu/zfsbootmenu-lib.sh

@@ -18,7 +18,7 @@ zlog() {
   [ "${1}" -le "${loglevel:=4}" ] || return
 
   # shellcheck disable=SC2086
-  _script="$( basename $0 )"
+  _script="$( basename -- $0 )"
   _func="${FUNCNAME[2]}"
 
   # Only add script/function tracing to debug messages
@@ -2095,6 +2095,7 @@ emergency_shell() {
 
   tput clear
   tput cnorm
+  stty echo
 
   echo -n "Launching emergency shell: "
   echo -e "${message}\n"
@@ -2120,6 +2121,62 @@ zbmcmdline() {
   [ -f "${BASE}/zbm.cmdline" ] && echo | cat "${BASE}/zbm.cmdline" -
 }
 
+# arg1: pid
+# prints: child pid
+# returns: 0 if a child pid was found, 1 if there are no children
+
+find_child_pid() {
+  local pid child
+
+  pid="${1}"
+  if [ -z "${pid}" ]; then
+    zdebug "empty pid"
+    return 1
+  fi
+
+  if [ -e "/proc/${pid}/task/${pid}/children" ] ; then
+    read -r child < "/proc/${pid}/task/${pid}/children"
+
+    if [ -n "${child}" ] ; then
+      echo "${child}"
+      return 0
+    fi
+  fi
+
+  return 1
+}
+
+# prints: nothing
+# returns: nothing
+
+takeover() {
+  local pid child
+
+  if [ -e "${BASE}/active" ] ; then
+    read -r pid < "${BASE}/active"
+    parent=${pid}
+
+    zinfo "Stopping active zfsbootmenu with a PID of ${pid}"
+
+    # Trip the USR1 handler in /bin/zfsbootmenu - 'exit 0'
+    zdebug "sending USR1 to ${parent}"
+    kill -USR1 "${parent}"
+
+    # find the last child process of the active /bin/zfsbootmenu
+    while child="$( find_child_pid "${parent}" )" ; do
+      if [ -n "${child}" ] ; then
+        parent=${child}
+        continue
+      fi
+      break
+    done
+
+    # Kill the blocking child so that the USR1 handler actually executes
+    zdebug "killing child process ${parent}"
+    [ -n "${parent}" ] && kill "${parent}"
+  fi
+}
+
 # prints: nothing
 # returns: 0
 

90zfsbootmenu/zfsbootmenu-preinit.sh

@@ -24,6 +24,7 @@ mkdir -p "${BASE}"
 # shellcheck disable=SC2154
 cat >> "/etc/profile" <<EOF
 # Added by zfsbootmenu-preinit.sh
+export BASE="/zfsbootmenu"
 export endian="${endian}"
 export spl_hostid="${spl_hostid}"
 export import_policy="${import_policy}"

90zfsbootmenu/zfsbootmenu.sh

@@ -19,10 +19,6 @@ fi
 # Make sure /dev/zfs exists, otherwise drop to a recovery shell
 [ -e /dev/zfs ] || emergency_shell "/dev/zfs missing, check that kernel modules are loaded"
 
-if [ -z "${BASE}" ]; then
-  export BASE="/zfsbootmenu"
-fi
-
 mkdir -p "${BASE}"
 
 while [ ! -e "${BASE}/initialized" ]; do
@@ -34,8 +30,11 @@ while [ ! -e "${BASE}/initialized" ]; do
   fi
 done
 
+[ -e "${BASE}/active" ] && takeover
+
+# If the takeover fails for some reason, spin until it ends
 while [ -e "${BASE}/active" ]; do
-  if ! delay=5 prompt="Press [ESC] to cancel" timed_prompt "Waiting for other ZFSBootMenu instace to terminate"; then
+  if ! delay=1 prompt="Press [ESC] to cancel" timed_prompt "Waiting for other ZFSBootMenu instance to terminate"; then
     zdebug "exited while waiting to own ${BASE}/active"
     tput cnorm
     tput clear
@@ -44,11 +43,13 @@ while [ -e "${BASE}/active" ]; do
 done
 
 # Prevent conflicting use of the boot menu
-: > "${BASE}/active"
+echo "$$" > "${BASE}/active"
 zdebug "creating ${BASE}/active"
 
 # shellcheck disable=SC2064
 trap "rm -f '${BASE}/active'" EXIT
+trap "zdebug 'exiting via USR1 signal' ; tput clear ; exit 0" SIGUSR1
+trap '' SIGINT
 
 if [ -r "${BASE}/bootfs" ]; then
   read -r BOOTFS < "${BASE}/bootfs"
@@ -66,8 +67,6 @@ if [ -d /libexec/setup.d ]; then
   unset _hook
 fi
 
-trap '' SIGINT
-
 # shellcheck disable=SC2016
 fuzzy_default_options=( "--ansi" "--no-clear"
   "--layout=reverse-list" "--inline-info" "--tac" "--color=16"