#!/bin/sh
# Auto-restart binaries with changed libraries
# Copyright 2010-2014 Gentoo Foundation; Distributed under the GPL v2
# Author: Christian Ruppert <idl0r@gentoo.org>

VERBOSE=0
REPORT=1
CHECK=0
FIX=0

get_ppids() {
	local pid=${1:-1}

	[ $pid -eq 1 ] && return

	local ppid="$(awk '$1 == "PPid:" { print $2 }' /proc/$pid/status 2>/dev/null || echo 0)"
	[ $ppid -eq 1 ] && return
	# "Error" case
	[ $ppid -eq 0 ] && return
	echo $ppid
	get_ppids $ppid
}

init_restart() {
	local service=$1

	[ ! -f /etc/init.d/$service ] && return 0

	# Only restart if it is really running through the init script
	if /etc/init.d/$service --quiet status 1>/dev/null 2>&1; then
		echo "Restarting ${service}..."
		/etc/init.d/$service restart
	elif [ -f /etc/debian_version ]; then
		echo "Restarting ${service}..."
		/etc/init.d/$service restart
	else
		local ret=${?:-0}

		# service crashed
		if [ ${?:-0} -eq 32 ]; then
			echo "Restarting crashed service: ${service} ..."
			/etc/init.d/$service restart
		elif [ $ret -eq 3 ]; then
			echo "It seems service \"${service}\" has been started \"by hand\""
			echo "would suggest to run \"killall ${service}; /etc/init.d/${service} start\""
		fi
	fi
}

deleted_files() {
	local mode=${1:-report}
	local only_if="${2}"
	local check_out=""
	local exit_code=0

	# deleted files test, lsof alternative
	for proc_pid in /proc/[0-9]*; do
		local pid=$(basename -- "${proc_pid}")
		# Skip init
		[ $pid -eq 1 ] && continue
		# Skip already fixed or gone procs
		[ ! -d "${proc_pid}" ] && continue

		local cmd=$(cut -d '' -f 1 $proc_pid/cmdline 2>/dev/null)

		[ -z "${cmd}" ] && continue

		local cmdline="$(cut -d '' --output-delimiter=" " -f 2- $proc_pid/cmdline)"

		# NOTE: "uniq" wouldn't work here
		local deleted="$(awk '$7 == "(deleted)" { print $6 }' $proc_pid/maps 2>/dev/null |egrep -v '^/SYSV[[:xdigit:]]+' | sort -u)"
		[ -z "${deleted}" ] && continue

		# Skip Threads
		local ppid="$(awk '$1 == "PPid:" { print $2 }' $proc_pid/status 2>/dev/null)"
		if [ -n "$ppid" -a "$ppid" -gt 1 ]; then
			local ppid_command=$(cut -d '' -f 1 /proc/$ppid/cmdline 2>/dev/null)
			[ "${cmd}" = "${ppid_command}" ] && continue
		fi
		local ppids="$(get_ppids $pid | sort -r | xargs)"
		local ppid_first="$(echo $ppids | awk 'BEGIN { FS = " "; } { print $1 }')"
		[ -z $ppid_first ] && ppid_first=1

		local command_base="$(basename -- "${cmd}" | awk '{ print $1 }')"
		local command_dir="$(dirname -- "${cmd}")"

		# Ignore /dev/zero, /var/run/, /run/, /SYS and /proc/, /[aio], /var/tmp/mysql
		[ -z "$(echo $deleted | awk '{ for (i = 1; i <= NF; i++) { if (!match($i, /^\/(dev\/zero|((var\/)?run|SYS|proc)\/|\[aio\]|var\/tmp\/mysql)/ )) print $i; } }')" ] && continue

		if [ -n "${only_if}" ]; then
			local skip=1
			for del in $deleted; do
				if [ -n "$(echo \"${del}\" | grep ${only_if})" ]; then
					skip=0
					break
				fi
			done
			[ $skip -eq 1 ] && continue
		fi

		# Anything beyond should actually return 1
		exit_code=1

		if [ "${mode}" = "report" -o "${mode}" = "check" ]; then
			case $cmd in
				# Ignore those
				SCREEN|sshd:*|su|sudo|/bin/login|ssh-agent|irssi )
					if [ $VERBOSE -eq 1 ]; then
						echo "Found a process which is still using deleted/updated files/libs:"
						echo "PID: ${pid}"
						[ -n "${ppids}" ] && echo "PPIDs: ${ppids}"
						echo "Command: ${cmd}"
						echo "Cmdline: ${cmdline}"
						echo "${deleted}"
						echo
						[ $ppid_first -gt 1 ] && pstree -naup $ppid_first
						echo
					fi
					exit_code=0
					continue
					;;
				* )
					if [ "${mode}" = "report" ]; then
						echo "Found a process which is still using deleted/updated files/libs:"
						echo "PID: ${pid}"
						[ -n "${ppids}" ] && echo "PPIDs: ${ppids}"
						echo "Command: ${cmd}"
						echo "Cmdline: ${cmdline}"
						echo "${deleted}"
						echo
						[ $ppid_first -gt 1 ] && pstree -naup $ppid_first
						echo
					fi
					check_out="${check_out:+${check_out}, }${command_base}(${pid})"
					continue
					;;
			esac
		elif [ "${mode}" = "fix" ]; then
			case $command_dir in
				/usr/sbin|/sbin|/usr/bin|/usr/local/bin|/usr/local/sbin )
					case $command_base in
						getty|agetty )
							pkill '^[a]getty$'
							continue
							;;
						cron )
							init_restart cronie
							continue
							;;
						rsync )
							# Evil rsync hack
							rsyncd_conf="$(awk 'BEGIN { FS = "\0"; } { for (i = 1; i < NF; i++) { if (match($i, /^--config=(.*\/)?([^\/]+)\.conf$/, m)) print m[2]; } }' ${proc_pid}/cmdline)"
							if [ -n "${rsyncd_conf}" ] && [ -f "/etc/init.d/${rsyncd_conf}" ]; then
								init_restart $rsyncd_conf
							else
								init_restart rsyncd
							fi
							continue
							;;
						apache2 )
							# Workaround the restart issues with the apache2 init script
							if [ -f /etc/gentoo_release ]; then
								/etc/init.d/apache2 stop; sleep 5; /etc/init.d/apache2 zap start
							else
								init_restart apache2
							fi
							continue
							;;
						mysqld )
							mysqld_conf="$(awk 'BEGIN { FS = "\0"; } { for (i = 1; i < NF; i++) { if (match($i, /^--defaults-file=(.*\/)?(my-?)?([^\/]+)\.cnf$/, m)) print m[3]; } }' ${proc_pid}/cmdline)"
							if [ -n "${mysqld_conf}" ] && [ -f "/etc/init.d/mysql-${mysqld_conf}" ]; then
								init_restart mysql-${mysqld_conf}
							else
								init_restart mysql
							fi
							continue
							;;
						flow-capture)
							init_restart flowcapture
							continue
							;;
						*)
							init_restart $command_base
							continue
							;;
					esac
					;;
				* )
					case $cmd in
						tlsmgr|qmgr|/usr/lib/postfix/master|/usr/libexec/postfix/master )
							init_restart postfix
							continue ;;
						/usr/lib*/postgresql-9.0/bin/postgres)
							init_restart postgresql-9.0
							continue ;;
						/usr/lib*/postgresql-8.4/bin/postgres)
							init_restart postgresql-8.4
							continue ;;
						# Workaround for "broken" proc titles with mod_perl/bugzilla
						/var/www/bugs.gen )
							# Workaround the restart issues with the apache2 init script
							/etc/init.d/apache2 stop; sleep 5; /etc/init.d/apache2 zap start
#							init_restart apache2
							continue ;;
						/var/www/bugs.gentoo.org/htdocs/jobqueue.pl )
							init_restart bugzilla-queue
							continue ;;
						*)
							init_restart $command_base
							continue
							;;
					esac
					;;
			esac

			# Try based on the whole cmdline (cmdline+1) as well
			# This special case is useful when a interpreter is used, like perl
			# or python.
			# NOTE: We may need to tweak that more later to use /proc/<pid>/cmdline[1] only
			if [ -n "${cmdline}" ]; then
				case $cmdline in
					/usr/local/sbin/apache-segfault-watch )
						kill $pid
						/usr/local/sbin/apache-segfault-watch
						continue
						;;
				esac
			fi
		fi
	done

	if [ "${mode}" = "check" ]; then
		if [ -n "${check_out}" ]; then
			echo "The following processes are using updated/deleted files: ${check_out}"
			exit_code=1
		else
			echo "No updated/deleted files are used by running processes"
			exit_code=0
		fi
	fi

	return $exit_code
}

while getopts d:vrcfh name; do
	case $name in
		d)
			DELETED=$OPTARG
			;;

		v)
			VERBOSE=1
			;;
		r )
			REPORT=1
			CHECK=0
			FIX=0
			;;
		c )
			CHECK=1
			REPORT=0
			FIX=0
			;;
		f )
			FIX=1
			CHECK=0
			REPORT=0
			;;
		?|h)
			printf "Usage: %s: [-v] [-r] [-c] [-d DELETED]\n" $0
			printf "\t-d DELETED - find/restart services still using the specified deleted file\n"
			printf "\t-v         - verbose during report, will also show otherwise ignored procs\n"
			printf "\t-r         - report only (default)\n"
			printf "\t-c         - check only, output format and exit values are meant to be used for Nagios or Icinga\n"
			printf "\t-f         - try to fix (restart) services\n"
			exit 2
			;;
	esac
done
shift $(($OPTIND - 1))

if [ $REPORT -eq 1 ]; then
    deleted_files "report" "${DELETED}"
elif [ $FIX -eq 1 ]; then
    deleted_files "fix" "${DELETED}"
    deleted_files "report" "${DELETED}"
elif [ $CHECK -eq 1 ]; then
    deleted_files "check" "${DELETED}"
fi

exit $?