#!/bin/bash
# Auto-restart binaries with changed libraries
# Copyright 2010-2012 Gentoo Foundation; Distributed under the GPL v2
# Author: Christian Ruppert <idl0r@gentoo.org>

VERBOSE=0
REPORT_ONLY=0

# lsof -n Do not resolve hostnames (no DNS).
# -P Do not resolve port names (list port number instead of its name).

# pstree -a -pu

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

	# Only restart if it is really running through the init script
	if /etc/init.d/$service --quiet status 2> /dev/null; then
		echo "Restarting ${service}..."
		/etc/init.d/$service restart
	else
		# service crashed
		if [ ${?:-0} -eq 32 ]; then
			echo "Restarting crashed service: ${service} ..."
			/etc/init.d/$service restart
		fi
	fi
}

deleted_files() {
	local report=${1:-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 command="$(awk 'BEGIN { FS = "\0"; } { print $1 }' $proc_pid/cmdline 2>/dev/null)"
		[ -z "${command}" ] && continue

		local cmdline="$(awk 'BEGIN { ORS = " "; FS = "\0"; } { for (i = 2; i < NF; i++) { print $i } }' $proc_pid/cmdline 2>/dev/null | sed 's/[[:space:]]$//g')"

		# 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 [ $ppid -gt 1 ]; then
			local ppid_command="$(awk 'BEGIN { FS = "\0"; } { print $1 }' /proc/$ppid/cmdline 2>/dev/null)"
			[ "${command}" = "${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="$(basename -- "${command}")"
		local command_dir="$(dirname -- "${command}")"

		if [ $report -eq 1 ]; then
			case $command 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: ${command}"
						echo "Cmdline: ${cmdline}"
						echo "${deleted}"
						echo
						[ $ppid_first -gt 1 ] && pstree -naup $ppid_first
						echo
					fi
					continue
					;;
				* )
					# Special hack for nscd, ignore "false-positives"
					if [ "${command}" = "/usr/sbin/nscd" ]; then
						[ -z "$(echo $deleted | awk '{ for (i = 1; i <= NF; i++) { if (!match($i, /^\/var\/run\/nscd\//)) print $i; } }')" ] && continue
					elif [ "${command}" = "/usr/sbin/apache2" -o "${command}" = "/var/www/bugs.gen" ]; then
						[ -z "$(echo $deleted | awk '{ for (i = 1; i <= NF; i++) { if (!match($i, /^\/(dev\/zero)/)) print $i; } }')" ] && continue
					fi

					echo "Found a process which is still using deleted/updated files/libs:"
					echo "PID: ${pid}"
					[ -n "${ppids}" ] && echo "PPIDs: ${ppids}"
					echo "Command: ${command}"
					echo "Cmdline: ${cmdline}"
					echo "${deleted}"
					echo
					[ $ppid_first -gt 1 ] && pstree -naup $ppid_first
					echo
					;;
			esac
		else
			case $command_dir in
				/usr/sbin|/sbin|/usr/bin|/usr/local/bin|/usr/local/sbin )
					case $command_base in
						atd|snmpd|sshd|multipathd|thrulayd|ntpd|fprobe|ulogd|syslog-ng|stunnel|xinetd|incrond|memcached|haproxy|cfservd|nrpe|named|auditd|dmeventd|smartd|nagios|varnishd|keepalived|dovecot )
							init_restart $command_base
							continue ;;
						agetty )
							pkill agetty
							continue
							;;
						cron )
							init_restart vixie-cron
							continue
							;;
						rsync )
							# Evil rsync hack
							rsyncd_conf="$(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 $rsyncd_conf
							else
								init_restart rsyncd
							fi
							continue
							;;
						nscd )
							# Special hack for nscd, ignore "false-positives"
							[ -z "$(echo $deleted | awk '{ for (i = 1; i <= NF; i++) { if (!match($i, /^\/var\/run\/nscd\//)) print $i; } }')" ] && continue
							init_restart nscd
							continue
							;;
						apache2 )
							[ -z "$(echo $deleted | awk '{ for (i = 1; i <= NF; i++) { if (!match($i, /^\/(dev\/zero)/)) print $i; } }')" ] && continue
							init_restart apache2
							continue
							;;
						mysqld )
							mysqld_conf="$(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
							;;
						bmc-watchdog)
							init_restart bmc-watchdog
							continue
							;;
					esac
					;;
				* )
					case $command in
						tlsmgr|qmgr|/usr/lib/postfix/master )
							init_restart postfix
							continue ;;
						mdadm )
							init_restart mdadm
							continue ;;
						nrpe )
							init_restart nrpe
							continue ;;
						/usr/lib*/openldap/slapd )
							init_restart slapd
							continue ;;
						/usr/lib*/postgresql-9.0/bin/postgres)
							init_restart postgresql-9.0
							continue ;;
						/usr/lib*/postgresql-8.4/bin/postgres)
							init_restart postgresql-8.4
							continue ;;
						# Workaround for "broken" proc titles with mod_perl/bugzilla
						/var/www/bugs.gen )
							[ -z "$(echo $deleted | awk '{ for (i = 1; i <= NF; i++) { if (!match($i, /^\/(dev\/zero)/)) print $i; } }')" ] && continue
							init_restart apache2
							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
}

while getopts hvr name; do
	case $name in
		v)
			VERBOSE=1 ;;
		r )
			REPORT_ONLY=1 ;;
		?|h)
			printf "Usage: %s: [-v] [-r]\n" $0
			printf "\t-v - verbose during report, will also show otherwise ignored procs\n"
			printf "\t-r - report only\n"
			exit 2 ;;
	esac
done
shift $(($OPTIND - 1))

[ $REPORT_ONLY -eq 0 ] && deleted_files 0
deleted_files 1