config/functions

   1 # Hey Emacs, this is a -*- shell-script -*- !!!
   2
   3 # utility functions for ctdb event scripts
   4
   5 [ -z "$CTDB_VARDIR" ] && {
   6     if [ -d "/var/lib/ctdb" ] ; then
   7         export CTDB_VARDIR="/var/lib/ctdb"
   8     else
   9         export CTDB_VARDIR="/var/ctdb"
  10     fi
  11 }
  12 [ -z "$CTDB_ETCDIR" ] && {
  13     export CTDB_ETCDIR="/etc"
  14 }
  15
  16 #######################################
  17 # pull in a system config file, if any
  18 _loadconfig() {
  19
  20     if [ -z "$1" ] ; then
  21         foo="${service_config:-${service_name}}"
  22         if [ -n "$foo" ] ; then
  23             loadconfig "$foo"
  24             return
  25         fi
  26     fi
  27
  28     if [ "$1" != "ctdb" ] ; then
  29         loadconfig "ctdb"
  30     fi
  31
  32     if [ -z "$1" ] ; then
  33         return
  34     fi
  35
  36     if [ -f $CTDB_ETCDIR/sysconfig/$1 ]; then
  37         . $CTDB_ETCDIR/sysconfig/$1
  38     elif [ -f $CTDB_ETCDIR/default/$1 ]; then
  39         . $CTDB_ETCDIR/default/$1
  40     elif [ -f $CTDB_BASE/sysconfig/$1 ]; then
  41         . $CTDB_BASE/sysconfig/$1
  42     fi
  43
  44     if [ "$1" = "ctdb" ] ; then
  45         _config="${CTDB_BASE}/ctdbd.conf"
  46         if [ -r "$_config" ] ; then
  47             . "$_config"
  48         fi
  49     fi
  50 }
  51
  52 loadconfig () {
  53     _loadconfig "$@"
  54 }
  55
  56 ##############################################################
  57
  58 # CTDB_SCRIPT_DEBUGLEVEL can be overwritten by setting it in a
  59 # configuration file.
  60 debug ()
  61 {
  62     if [ ${CTDB_SCRIPT_DEBUGLEVEL:-2} -ge 4 ] ; then
  63         # If there are arguments then echo them.  Otherwise expect to
  64         # use stdin, which allows us to pass lots of debug using a
  65         # here document.
  66         if [ -n "$1" ] ; then
  67             echo "DEBUG: $*"
  68         elif ! tty -s ; then
  69             sed -e 's@^@DEBUG: @'
  70         fi
  71     fi
  72 }
  73
  74 die ()
  75 {
  76     _msg="$1"
  77     _rc="${2:-1}"
  78
  79     echo "$_msg"
  80     exit $_rc
  81 }
  82
  83 # Log given message or stdin to either syslog or a CTDB log file
  84 # $1 is the tag passed to logger if syslog is in use.
  85 script_log ()
  86 {
  87     _tag="$1" ; shift
  88
  89     if [ "$CTDB_SYSLOG" = "yes" ] ; then
  90         logger -t "ctdbd: ${_tag}" $*
  91     else
  92         {
  93             if [ -n "$*" ] ; then
  94                 echo "$*"
  95             else
  96                 cat
  97             fi
  98         } >>"${CTDB_LOGFILE:-/var/log/log.ctdb}"
  99     fi
 100 }
 101
 102 # When things are run in the background in an eventscript then logging
 103 # output might get lost.  This is the "solution".  :-)
 104 background_with_logging ()
 105 {
 106     (
 107         "$@" 2>&1 </dev/null |
 108         script_log "${script_name}&"
 109     )&
 110
 111     return 0
 112 }
 113
 114 ##############################################################
 115 # check number of args for different events
 116 ctdb_check_args ()
 117 {
 118     case "$1" in
 119         takeip|releaseip)
 120             if [ $# != 4 ]; then
 121                 echo "ERROR: must supply interface, IP and maskbits"
 122                 exit 1
 123             fi
 124             ;;
 125         updateip)
 126             if [ $# != 5 ]; then
 127                 echo "ERROR: must supply old interface, new interface, IP and maskbits"
 128                 exit 1
 129             fi
 130             ;;
 131     esac
 132 }
 133
 134 ##############################################################
 135 # determine on what type of system (init style) we are running
 136 detect_init_style()
 137 {
 138     # only do detection if not already set:
 139     [ -z "$CTDB_INIT_STYLE" ] || return
 140
 141     if [ -x /sbin/startproc ]; then
 142         CTDB_INIT_STYLE="suse"
 143     elif [ -x /sbin/start-stop-daemon ]; then
 144         CTDB_INIT_STYLE="debian"
 145     else
 146         CTDB_INIT_STYLE="redhat"
 147     fi
 148 }
 149
 150 ######################################################
 151 # simulate /sbin/service on platforms that don't have it
 152 # _service() makes it easier to hook the service() function for
 153 # testing.
 154 _service ()
 155 {
 156   _service_name="$1"
 157   _op="$2"
 158
 159   # do nothing, when no service was specified
 160   [ -z "$_service_name" ] && return
 161
 162   if [ -x /sbin/service ]; then
 163       $_nice /sbin/service "$_service_name" "$_op"
 164   elif [ -x $CTDB_ETCDIR/init.d/$_service_name ]; then
 165       $_nice $CTDB_ETCDIR/init.d/$_service_name "$_op"
 166   elif [ -x $CTDB_ETCDIR/rc.d/init.d/$_service_name ]; then
 167       $_nice $CTDB_ETCDIR/rc.d/init.d/$_service_name "$_op"
 168   fi
 169 }
 170
 171 service()
 172 {
 173     _nice=""
 174     _service "$@"
 175 }
 176
 177 ######################################################
 178 # simulate /sbin/service (niced) on platforms that don't have it
 179 nice_service()
 180 {
 181     _nice="nice"
 182     _service "$@"
 183 }
 184
 185 ######################################################
 186 # wrapper around /proc/ settings to allow them to be hooked
 187 # for testing
 188 # 1st arg is relative path under /proc/, 2nd arg is value to set
 189 set_proc ()
 190 {
 191     echo "$2" >"/proc/$1"
 192 }
 193
 194 ######################################################
 195 # wrapper around getting file contents from /proc/ to allow
 196 # this to be hooked for testing
 197 # 1st arg is relative path under /proc/
 198 get_proc ()
 199 {
 200     cat "/proc/$1"
 201 }
 202
 203 ######################################################
 204 # Check that an RPC service is healthy -
 205 # this includes allowing a certain number of failures
 206 # before marking the NFS service unhealthy.
 207 #
 208 # usage: nfs_check_rpc_service SERVICE_NAME [ triple ...]
 209 #
 210 # each triple is a set of 3 arguments: an operator, a
 211 # fail count limit and an action string.
 212 #
 213 # For example:
 214 #
 215 #       nfs_check_rpc_service "lockd" \
 216 #           -ge 15 "verbose restart unhealthy" \
 217 #           -eq 10 "restart:bs"
 218 #
 219 # says that if lockd is down for 15 iterations then do
 220 # a verbose restart of lockd and mark the node unhealthy.
 221 # Before this, after 10 iterations of failure, the
 222 # service is restarted silently in the background.
 223 # Order is important: the number of failures need to be
 224 # specified in reverse order because processing stops
 225 # after the first condition that is true.
 226 ######################################################
 227 nfs_check_rpc_service ()
 228 {
 229     _prog_name="$1" ; shift
 230
 231     if _nfs_check_rpc_common "$_prog_name" ; then
 232         return
 233     fi
 234
 235     while [ -n "$3" ] ; do
 236         if _nfs_check_rpc_action "$1" "$2" "$3" ; then
 237             break
 238         fi
 239         shift 3
 240     done
 241 }
 242
 243 # The new way of doing things...
 244 nfs_check_rpc_services ()
 245 {
 246     # Files must end with .check - avoids editor backups, RPM fu, ...
 247     for _f in "${CTDB_BASE}/nfs-rpc-checks.d/"[0-9][0-9].*.check ; do
 248         _t="${_f%.check}"
 249         _prog_name="${_t##*/[0-9][0-9].}"
 250
 251         if _nfs_check_rpc_common "$_prog_name" ; then
 252             # This RPC service is up, check next service...
 253             continue
 254         fi
 255
 256         # Check each line in the file in turn until one of the limit
 257         # checks is hit...
 258         while read _cmp _lim _rest ; do
 259             # Skip comments
 260             case "$_cmp" in
 261                 \#*) continue ;;
 262             esac
 263
 264             if _nfs_check_rpc_action "$_cmp" "$_lim" "$_rest" ; then
 265                 # Limit was hit on this line, no further checking...
 266                 break
 267             fi
 268         done <"$_f"
 269     done
 270 }
 271
 272 _nfs_check_rpc_common ()
 273 {
 274     _prog_name="$1"
 275
 276     # Some platforms don't have separate programs for all services.
 277     case "$_prog_name" in
 278         statd)
 279             which "rpc.${_prog_name}" >/dev/null 2>&1 || return 0
 280     esac
 281
 282     case "$_prog_name" in
 283         nfsd)
 284             _rpc_prog=nfs
 285             _version=3
 286             ;;
 287         mountd)
 288             _rpc_prog=mountd
 289             _version=1
 290             ;;
 291         rquotad)
 292             _rpc_prog=rquotad
 293             _version=1
 294             ;;
 295         lockd)
 296             _rpc_prog=nlockmgr
 297             _version=4
 298             ;;
 299         statd)
 300             _rpc_prog=status
 301             _version=1
 302             ;;
 303         *)
 304             echo "Internal error: unknown RPC program \"$_prog_name\"."
 305             exit 1
 306     esac
 307
 308     _service_name="nfs_${_prog_name}"
 309
 310     if ctdb_check_rpc "$_rpc_prog" $_version >/dev/null ; then
 311         ctdb_counter_init "$_service_name"
 312         return 0
 313     fi
 314
 315     ctdb_counter_incr "$_service_name"
 316
 317     return 1
 318 }
 319
 320 _nfs_check_rpc_action ()
 321 {
 322     _cmp="$1"
 323     _limit="$2"
 324     _actions="$3"
 325
 326     if ctdb_check_counter "quiet" "$_cmp" "$_limit" "$_service_name" ; then
 327         return 1
 328     fi
 329
 330     for _action in $_actions ; do
 331         case "$_action" in
 332             verbose)
 333                 echo "$ctdb_check_rpc_out"
 334                 ;;
 335             restart)
 336                 _nfs_restart_rpc_service "$_prog_name"
 337                 ;;
 338             restart:b)
 339                 _nfs_restart_rpc_service "$_prog_name" true
 340                 ;;
 341             unhealthy)
 342                 exit 1
 343                 ;;
 344             *)
 345                 echo "Internal error: unknown action \"$_action\"."
 346                 exit 1
 347         esac
 348     done
 349
 350     return 0
 351 }
 352
 353 _nfs_restart_rpc_service ()
 354 {
 355     _prog_name="$1"
 356     _background="${2:-false}"
 357
 358     if $_background ; then
 359         _maybe_background="background_with_logging"
 360     else
 361         _maybe_background=""
 362     fi
 363
 364     _p="rpc.${_prog_name}"
 365
 366     case "$_prog_name" in
 367         nfsd)
 368             echo "Trying to restart NFS service"
 369             $_maybe_background startstop_nfs restart
 370             ;;
 371         mountd)
 372             echo "Trying to restart $_prog_name [${_p}]"
 373             killall -q -9 "$_p"
 374             $_maybe_background $_p ${MOUNTD_PORT:+-p} $MOUNTD_PORT
 375             ;;
 376         rquotad)
 377             echo "Trying to restart $_prog_name [${_p}]"
 378             killall -q -9 "$_p"
 379             $_maybe_background $_p ${RQUOTAD_PORT:+-p} $RQUOTAD_PORT
 380             ;;
 381         lockd)
 382             echo "Trying to restart lock manager service"
 383             $_maybe_background startstop_nfslock restart
 384             ;;
 385         statd)
 386             echo "Trying to restart $_prog_name [${_p}]"
 387             killall -q -9 "$_p"
 388             $_maybe_background $_p \
 389                 ${STATD_HOSTNAME:+-n} $STATD_HOSTNAME \
 390                 ${STATD_PORT:+-p} $STATD_PORT \
 391                 ${STATD_OUTGOING_PORT:+-o} $STATD_OUTGOING_PORT
 392             ;;
 393         *)
 394             echo "Internal error: unknown RPC program \"$_prog_name\"."
 395             exit 1
 396     esac
 397 }
 398
 399 ######################################################
 400 # check that a rpc server is registered with portmap
 401 # and responding to requests
 402 # usage: ctdb_check_rpc SERVICE_NAME VERSION
 403 ######################################################
 404 ctdb_check_rpc ()
 405 {
 406     progname="$1"
 407     version="$2"
 408
 409     _localhost="${CTDB_RPCINFO_LOCALHOST:-127.0.0.1}"
 410
 411     if ! ctdb_check_rpc_out=$(rpcinfo -u $_localhost $progname $version 2>&1) ; then
 412         ctdb_check_rpc_out="ERROR: $progname failed RPC check:
 413 $ctdb_check_rpc_out"
 414         echo "$ctdb_check_rpc_out"
 415         return 1
 416     fi
 417 }
 418
 419 ######################################################
 420 # Ensure $service_name is set
 421 assert_service_name ()
 422 {
 423     [ -n "$service_name" ] || die "INTERNAL ERROR: \$service_name not set"
 424 }
 425
 426 ######################################################
 427 # check a set of directories is available
 428 # return 1 on a missing directory
 429 # directories are read from stdin
 430 ######################################################
 431 ctdb_check_directories_probe()
 432 {
 433     while IFS="" read d ; do
 434         case "$d" in
 435             *%*)
 436                 continue
 437                 ;;
 438             *)
 439                 [ -d "${d}/." ] || return 1
 440         esac
 441     done
 442 }
 443
 444 ######################################################
 445 # check a set of directories is available
 446 # directories are read from stdin
 447 ######################################################
 448 ctdb_check_directories()
 449 {
 450     ctdb_check_directories_probe || {
 451         echo "ERROR: $service_name directory \"$d\" not available"
 452         exit 1
 453     }
 454 }
 455
 456 ######################################################
 457 # check a set of tcp ports
 458 # usage: ctdb_check_tcp_ports <ports...>
 459 ######################################################
 460
 461 # This flag file is created when a service is initially started.  It
 462 # is deleted the first time TCP port checks for that service succeed.
 463 # Until then ctdb_check_tcp_ports() prints a more subtle "error"
 464 # message if a port check fails.
 465 _ctdb_check_tcp_common ()
 466 {
 467     assert_service_name
 468     _ctdb_service_started_file="$ctdb_fail_dir/$service_name.started"
 469 }
 470
 471 ctdb_check_tcp_init ()
 472 {
 473     _ctdb_check_tcp_common
 474     mkdir -p "${_ctdb_service_started_file%/*}" # dirname
 475     touch "$_ctdb_service_started_file"
 476 }
 477
 478 # Check whether something is listening on all of the given TCP ports
 479 # using the "ctdb checktcpport" command.
 480 ctdb_check_tcp_ports()
 481 {
 482     if [ -z "$1" ] ; then
 483         echo "INTERNAL ERROR: ctdb_check_tcp_ports - no ports specified"
 484         exit 1
 485     fi
 486
 487     for _p ; do  # process each function argument (port)
 488         _cmd="ctdb checktcpport $_p"
 489         _out=$($_cmd 2>&1)
 490         _ret=$?
 491         case "$_ret" in
 492             0)
 493                 _ctdb_check_tcp_common
 494                 if [ ! -f "$_ctdb_service_started_file" ] ; then
 495                     echo "ERROR: $service_name tcp port $_p is not responding"
 496                     debug "\"ctdb checktcpport $_p\" was able to bind to port"
 497                 else
 498                     echo "INFO: $service_name tcp port $_p is not responding"
 499                 fi
 500
 501                 return 1
 502                 ;;
 503             98)
 504                 # Couldn't bind, something already listening, next port...
 505                 continue
 506                 ;;
 507             *)
 508                 echo "ERROR: unexpected error running \"ctdb checktcpport\""
 509                 debug <<EOF
 510 ctdb checktcpport (exited with $_ret) with output:
 511 $_out"
 512 EOF
 513                 return $_ret
 514         esac
 515     done
 516
 517     # All ports listening
 518     _ctdb_check_tcp_common
 519     rm -f "$_ctdb_service_started_file"
 520     return 0
 521 }
 522
 523 ######################################################
 524 # check a unix socket
 525 # usage: ctdb_check_unix_socket SERVICE_NAME <socket_path>
 526 ######################################################
 527 ctdb_check_unix_socket() {
 528     socket_path="$1"
 529     [ -z "$socket_path" ] && return
 530
 531     if ! netstat --unix -a -n | grep -q "^unix.*LISTEN.*${socket_path}$"; then
 532         echo "ERROR: $service_name socket $socket_path not found"
 533         return 1
 534     fi
 535 }
 536
 537 ######################################################
 538 # check a command returns zero status
 539 # usage: ctdb_check_command <command>
 540 ######################################################
 541 ctdb_check_command ()
 542 {
 543     _out=$("$@" 2>&1) || {
 544         echo "ERROR: $* returned error"
 545         echo "$_out" | debug
 546         exit 1
 547     }
 548 }
 549
 550 ################################################
 551 # kill off any TCP connections with the given IP
 552 ################################################
 553 kill_tcp_connections ()
 554 {
 555     _ip="$1"
 556
 557     _oneway=false
 558     if [ "$2" = "oneway" ] ; then
 559         _oneway=true
 560     fi
 561
 562     get_tcp_connections_for_ip "$_ip" | {
 563         _killcount=0
 564         _connections=""
 565         _nl="
 566 "
 567         while read _dst _src; do
 568             _destport="${_dst##*:}"
 569             __oneway=$_oneway
 570             case $_destport in
 571                 # we only do one-way killtcp for CIFS
 572                 139|445) __oneway=true ;;
 573             esac
 574
 575             echo "Killing TCP connection $_src $_dst"
 576             _connections="${_connections}${_nl}${_src} ${_dst}"
 577             if ! $__oneway ; then
 578                 _connections="${_connections}${_nl}${_dst} ${_src}"
 579             fi
 580
 581             _killcount=$(($_killcount + 1))
 582         done
 583
 584         if [ $_killcount -eq 0 ] ; then
 585             return
 586         fi
 587
 588         echo "$_connections" | ctdb killtcp || {
 589             echo "Failed to send killtcp control"
 590             return
 591         }
 592
 593         _count=0
 594         while : ; do
 595             _remaining=$(get_tcp_connections_for_ip $_ip | wc -l)
 596
 597             if [ $_remaining -eq 0 ] ; then
 598                 echo "Killed $_killcount TCP connections to released IP $_ip"
 599                 return
 600             fi
 601
 602             _count=$(($_count + 1))
 603             if [ $_count -gt 3 ] ; then
 604                 echo "Timed out killing tcp connections for IP $_ip ($_remaining remaining)"
 605                 return
 606             fi
 607
 608             echo "Waiting for $_remaining connections to be killed for IP $_ip"
 609             sleep 1
 610         done
 611     }
 612 }
 613
 614 ##################################################################
 615 # kill off the local end for any TCP connections with the given IP
 616 ##################################################################
 617 kill_tcp_connections_local_only ()
 618 {
 619     kill_tcp_connections "$1" "oneway"
 620 }
 621
 622 ##################################################################
 623 # tickle any TCP connections with the given IP
 624 ##################################################################
 625 tickle_tcp_connections ()
 626 {
 627     _ip="$1"
 628
 629     get_tcp_connections_for_ip "$_ip" |
 630     {
 631         _failed=false
 632
 633         while read dest src; do
 634             echo "Tickle TCP connection $src $dest"
 635             ctdb tickle $src $dest >/dev/null 2>&1 || _failed=true
 636             echo "Tickle TCP connection $dest $src"
 637             ctdb tickle $dest $src >/dev/null 2>&1 || _failed=true
 638         done
 639
 640         if $_failed ; then
 641             echo "Failed to send tickle control"
 642         fi
 643     }
 644 }
 645
 646 get_tcp_connections_for_ip ()
 647 {
 648     _ip="$1"
 649
 650     netstat -tn | awk -v ip=$_ip \
 651         'index($1, "tcp") == 1 && \
 652          (index($4, ip ":") == 1 || index($4, "::ffff:" ip ":") == 1) \
 653          && $6 == "ESTABLISHED" \
 654          {print $4" "$5}'
 655 }
 656
 657 ########################################################
 658 # start/stop the Ganesha nfs service
 659 ########################################################
 660 startstop_ganesha()
 661 {
 662     _service_name="nfs-ganesha-$CTDB_CLUSTER_FILESYSTEM_TYPE"
 663     case "$1" in
 664         start)
 665             service "$_service_name" start
 666             ;;
 667         stop)
 668             service "$_service_name" stop
 669             ;;
 670         restart)
 671             service "$_service_name" restart
 672             ;;
 673     esac
 674 }
 675
 676 ########################################################
 677 # start/stop the nfs service on different platforms
 678 ########################################################
 679 startstop_nfs() {
 680         PLATFORM="unknown"
 681         [ -x $CTDB_ETCDIR/init.d/nfsserver ] && {
 682                 PLATFORM="sles"
 683         }
 684         [ -x $CTDB_ETCDIR/init.d/nfslock ] && {
 685                 PLATFORM="rhel"
 686         }
 687
 688         case $PLATFORM in
 689         sles)
 690                 case $1 in
 691                 start)
 692                         service nfsserver start
 693                         ;;
 694                 stop)
 695                         service nfsserver stop > /dev/null 2>&1
 696                         ;;
 697                 restart)
 698                         set_proc "fs/nfsd/threads" 0
 699                         service nfsserver stop > /dev/null 2>&1
 700                         pkill -9 nfsd
 701                         nfs_dump_some_threads
 702                         service nfsserver start
 703                         ;;
 704                 esac
 705                 ;;
 706         rhel)
 707                 case $1 in
 708                 start)
 709                         service nfslock start
 710                         service nfs start
 711                         ;;
 712                 stop)
 713                         service nfs stop
 714                         service nfslock stop
 715                         ;;
 716                 restart)
 717                         set_proc "fs/nfsd/threads" 0
 718                         service nfs stop > /dev/null 2>&1
 719                         service nfslock stop > /dev/null 2>&1
 720                         pkill -9 nfsd
 721                         nfs_dump_some_threads
 722                         service nfslock start
 723                         service nfs start
 724                         ;;
 725                 esac
 726                 ;;
 727         *)
 728                 echo "Unknown platform. NFS is not supported with ctdb"
 729                 exit 1
 730                 ;;
 731         esac
 732 }
 733
 734 # Dump up to the configured number of nfsd thread backtraces.
 735 nfs_dump_some_threads ()
 736 {
 737     [ -n "$CTDB_NFS_DUMP_STUCK_THREADS" ] || return 0
 738
 739     # Optimisation to avoid running an unnecessary pidof
 740     [ $CTDB_NFS_DUMP_STUCK_THREADS -gt 0 ] || return 0
 741
 742     _count=0
 743     for _pid in $(pidof nfsd) ; do
 744         [ $_count -le $CTDB_NFS_DUMP_STUCK_THREADS ] || break
 745
 746         # Do this first to avoid racing with thread exit
 747         _stack=$(get_proc "${_pid}/stack" 2>/dev/null)
 748         if [ -n "$_stack" ] ; then
 749             echo "Stack trace for stuck nfsd thread [${_pid}]:"
 750             echo "$_stack"
 751             _count=$(($_count + 1))
 752         fi
 753     done
 754 }
 755
 756 ########################################################
 757 # start/stop the nfs lockmanager service on different platforms
 758 ########################################################
 759 startstop_nfslock() {
 760         PLATFORM="unknown"
 761         [ -x $CTDB_ETCDIR/init.d/nfsserver ] && {
 762                 PLATFORM="sles"
 763         }
 764         [ -x $CTDB_ETCDIR/init.d/nfslock ] && {
 765                 PLATFORM="rhel"
 766         }
 767
 768         case $PLATFORM in
 769         sles)
 770                 # for sles there is no service for lockmanager
 771                 # so we instead just shutdown/restart nfs
 772                 case $1 in
 773                 start)
 774                         service nfsserver start
 775                         ;;
 776                 stop)
 777                         service nfsserver stop > /dev/null 2>&1
 778                         ;;
 779                 restart)
 780                         service nfsserver stop > /dev/null 2>&1
 781                         service nfsserver start
 782                         ;;
 783                 esac
 784                 ;;
 785         rhel)
 786                 case $1 in
 787                 start)
 788                         service nfslock start
 789                         ;;
 790                 stop)
 791                         service nfslock stop > /dev/null 2>&1
 792                         ;;
 793                 restart)
 794                         service nfslock stop > /dev/null 2>&1
 795                         service nfslock start
 796                         ;;
 797                 esac
 798                 ;;
 799         *)
 800                 echo "Unknown platform. NFS locking is not supported with ctdb"
 801                 exit 1
 802                 ;;
 803         esac
 804 }
 805
 806 # Periodically update the statd database
 807 nfs_statd_update ()
 808 {
 809     _update_period="$1"
 810
 811     _statd_update_trigger="$service_state_dir/update-trigger"
 812     [ -f "$_statd_update_trigger" ] || touch "$_statd_update_trigger"
 813
 814     _last_update=$(stat --printf="%Y" "$_statd_update_trigger")
 815     _current_time=$(date +"%s")
 816     if [ $(( $_current_time - $_last_update)) -ge $_update_period ] ; then
 817         touch "$_statd_update_trigger"
 818         $CTDB_BASE/statd-callout updatelocal &
 819         $CTDB_BASE/statd-callout updateremote &
 820     fi
 821 }
 822
 823 add_ip_to_iface()
 824 {
 825     _iface=$1
 826     _ip=$2
 827     _maskbits=$3
 828
 829     _lockfile="${CTDB_VARDIR}/state/interface_modify_${_iface}.flock"
 830     mkdir -p "${_lockfile%/*}" # dirname
 831     [ -f "$_lockfile" ] || touch "$_lockfile"
 832
 833     (
 834         # Note: use of return/exit/die() below only gets us out of the
 835         # sub-shell, which is actually what we want.  That is, the
 836         # function should just return non-zero.
 837
 838         flock --timeout 30 0 || \
 839             die "add_ip_to_iface: unable to get lock for ${_iface}"
 840
 841         # Ensure interface is up
 842         ip link set "$_iface" up || \
 843             die "Failed to bringup interface $_iface"
 844
 845         ip addr add "$_ip/$_maskbits" brd + dev "$_iface" || \
 846             die "Failed to add $_ip/$_maskbits on dev $_iface"
 847     ) <"$_lockfile"
 848
 849     # Do nothing here - return above only gets us out of the subshell
 850     # and doing anything here will affect the return code.
 851 }
 852
 853 delete_ip_from_iface()
 854 {
 855     _iface=$1
 856     _ip=$2
 857     _maskbits=$3
 858
 859     _lockfile="${CTDB_VARDIR}/state/interface_modify_${_iface}.flock"
 860     mkdir -p "${_lockfile%/*}" # dirname
 861     [ -f "$_lockfile" ] || touch "$_lockfile"
 862
 863     (
 864         # Note: use of return/exit/die() below only gets us out of the
 865         # sub-shell, which is actually what we want.  That is, the
 866         # function should just return non-zero.
 867
 868         flock --timeout 30 0 || \
 869             die "delete_ip_from_iface: unable to get lock for ${_iface}"
 870
 871         _im="$_ip/$_maskbits"  # shorthand for readability
 872
 873         # "ip addr del" will delete all secondary IPs if this is the
 874         # primary.  To work around this _very_ annoying behaviour we
 875         # have to keep a record of the secondaries and re-add them
 876         # afterwards.  Yuck!
 877
 878         _secondaries=""
 879         if ip addr list dev "$_iface" primary | grep -Fq "inet $_im " ; then
 880             _secondaries=$(ip addr list dev "$_iface" secondary | \
 881                 awk '$1 == "inet" { print $2 }')
 882         fi
 883
 884         local _rc=0
 885         ip addr del "$_im" dev "$_iface" || {
 886             echo "Failed to del $_ip on dev $_iface"
 887             _rc=1
 888         }
 889
 890         if [ -n "$_secondaries" ] ; then
 891             for _i in $_secondaries; do
 892                 if ip addr list dev "$_iface" | grep -Fq "inet $_i" ; then
 893                     echo "Kept secondary $_i on dev $_iface"
 894                 else
 895                     echo "Re-adding secondary address $_i to dev $_iface"
 896                     ip addr add $_i brd + dev $_iface || {
 897                         echo "Failed to re-add address $_i to dev $_iface"
 898                         _rc=1
 899                     }
 900                 fi
 901             done
 902         fi
 903
 904         return $_rc
 905     ) <"$_lockfile"
 906
 907     # Do nothing here - return above only gets us out of the subshell
 908     # and doing anything here will affect the return code.
 909 }
 910
 911 # If the given IP is hosted then print 2 items: maskbits and iface
 912 ip_maskbits_iface ()
 913 {
 914     _addr="$1"
 915
 916     ip addr show to "${_addr}/32" 2>/dev/null | \
 917         awk '$1 == "inet" { print gensub(".*/", "", 1, $2), $NF }'
 918 }
 919
 920 drop_ip ()
 921 {
 922     _addr="${1%/*}"  # Remove optional maskbits
 923
 924     set -- $(ip_maskbits_iface $_addr)
 925     if [ -n "$1" ] ; then
 926         _maskbits="$1"
 927         _iface="$2"
 928         echo "Removing public address $_addr/$_maskbits from device $_iface"
 929         delete_ip_from_iface $_iface $_addr $_maskbits >/dev/null 2>&1
 930     fi
 931 }
 932
 933 drop_all_public_ips ()
 934 {
 935     while read _ip _x ; do
 936         drop_ip "$_ip"
 937     done <"${CTDB_PUBLIC_ADDRESSES:-/dev/null}"
 938 }
 939
 940 ########################################################
 941 # Simple counters
 942 _ctdb_counter_common () {
 943     _service_name="${1:-${service_name:-${script_name}}}"
 944     _counter_file="$ctdb_fail_dir/$_service_name"
 945     mkdir -p "${_counter_file%/*}" # dirname
 946 }
 947 ctdb_counter_init () {
 948     _ctdb_counter_common "$1"
 949
 950     >"$_counter_file"
 951 }
 952 ctdb_counter_incr () {
 953     _ctdb_counter_common "$1"
 954
 955     # unary counting!
 956     echo -n 1 >> "$_counter_file"
 957 }
 958 ctdb_check_counter () {
 959     _msg="${1:-error}"  # "error"  - anything else is silent on fail
 960     _op="${2:--ge}"  # an integer operator supported by test
 961     _limit="${3:-${service_fail_limit}}"
 962     shift 3
 963     _ctdb_counter_common "$1"
 964
 965     # unary counting!
 966     _size=$(stat -c "%s" "$_counter_file" 2>/dev/null || echo 0)
 967     _hit=false
 968     if [ "$_op" != "%" ] ; then
 969         if [ $_size $_op $_limit ] ; then
 970             _hit=true
 971         fi
 972     else
 973         if [ $(($_size $_op $_limit)) -eq 0 ] ; then
 974             _hit=true
 975         fi
 976     fi
 977     if $_hit ; then
 978         if [ "$_msg" = "error" ] ; then
 979             echo "ERROR: $_size consecutive failures for $_service_name, marking node unhealthy"
 980             exit 1
 981         else
 982             return 1
 983         fi
 984     fi
 985 }
 986
 987 ########################################################
 988
 989 ctdb_status_dir="$CTDB_VARDIR/status"
 990 ctdb_fail_dir="$CTDB_VARDIR/failcount"
 991
 992 ctdb_setup_service_state_dir ()
 993 {
 994     service_state_dir="$CTDB_VARDIR/state/${1:-${service_name}}"
 995     mkdir -p "$service_state_dir" || {
 996         echo "Error creating state dir \"$service_state_dir\""
 997         exit 1
 998     }
 999 }
1000
1001 ########################################################
1002 # Managed status history, for auto-start/stop
1003
1004 ctdb_managed_dir="$CTDB_VARDIR/managed_history"
1005
1006 _ctdb_managed_common ()
1007 {
1008     _ctdb_managed_file="$ctdb_managed_dir/$service_name"
1009 }
1010
1011 ctdb_service_managed ()
1012 {
1013     _ctdb_managed_common
1014     mkdir -p "$ctdb_managed_dir"
1015     touch "$_ctdb_managed_file"
1016 }
1017
1018 ctdb_service_unmanaged ()
1019 {
1020     _ctdb_managed_common
1021     rm -f "$_ctdb_managed_file"
1022 }
1023
1024 is_ctdb_previously_managed_service ()
1025 {
1026     _ctdb_managed_common
1027     [ -f "$_ctdb_managed_file" ]
1028 }
1029
1030 ########################################################
1031 # Check and set status
1032
1033 log_status_cat ()
1034 {
1035     echo "node is \"$1\", \"${script_name}\" reports problem: $(cat $2)"
1036 }
1037
1038 ctdb_checkstatus ()
1039 {
1040     if [ -r "$ctdb_status_dir/$script_name/unhealthy" ] ; then
1041         log_status_cat "unhealthy" "$ctdb_status_dir/$script_name/unhealthy"
1042         return 1
1043     elif [ -r "$ctdb_status_dir/$script_name/banned" ] ; then
1044         log_status_cat "banned" "$ctdb_status_dir/$script_name/banned"
1045         return 2
1046     else
1047         return 0
1048     fi
1049 }
1050
1051 ctdb_setstatus ()
1052 {
1053     d="$ctdb_status_dir/$script_name"
1054     case "$1" in
1055         unhealthy|banned)
1056             mkdir -p "$d"
1057             cat "$2" >"$d/$1"
1058             ;;
1059         *)
1060             for i in "banned" "unhealthy" ; do
1061                 rm -f "$d/$i"
1062             done
1063             ;;
1064     esac
1065 }
1066
1067 ##################################################################
1068 # Reconfigure a service on demand
1069
1070 _ctdb_service_reconfigure_common ()
1071 {
1072     _d="$ctdb_status_dir/${service_name}"
1073     mkdir -p "$_d"
1074     _ctdb_service_reconfigure_flag="$_d/reconfigure"
1075 }
1076
1077 ctdb_service_needs_reconfigure ()
1078 {
1079     _ctdb_service_reconfigure_common
1080     [ -e "$_ctdb_service_reconfigure_flag" ]
1081 }
1082
1083 ctdb_service_set_reconfigure ()
1084 {
1085     _ctdb_service_reconfigure_common
1086     >"$_ctdb_service_reconfigure_flag"
1087 }
1088
1089 ctdb_service_unset_reconfigure ()
1090 {
1091     _ctdb_service_reconfigure_common
1092     rm -f "$_ctdb_service_reconfigure_flag"
1093 }
1094
1095 ctdb_service_reconfigure ()
1096 {
1097     echo "Reconfiguring service \"${service_name}\"..."
1098     ctdb_service_unset_reconfigure
1099     service_reconfigure || return $?
1100     ctdb_counter_init
1101 }
1102
1103 # Default service_reconfigure() function does nothing.
1104 service_reconfigure ()
1105 {
1106     :
1107 }
1108
1109 ctdb_reconfigure_take_lock ()
1110 {
1111     _ctdb_service_reconfigure_common
1112     _lock="${_d}/reconfigure_lock"
1113     mkdir -p "${_lock%/*}" # dirname
1114     touch "$_lock"
1115
1116     (
1117         flock 0
1118         # This is overkill but will work if we need to extend this to
1119         # allow certain events to run multiple times in parallel
1120         # (e.g. takeip) and write multiple PIDs to the file.
1121         read _locker_event
1122         if [ -n "$_locker_event" ] ; then
1123             while read _pid ; do
1124                 if [ -n "$_pid" -a "$_pid" != $$ ] && \
1125                     kill -0 "$_pid" 2>/dev/null ; then
1126                     exit 1
1127                 fi
1128             done
1129         fi
1130
1131         printf "%s\n%s\n" "$event_name" $$ >"$_lock"
1132         exit 0
1133     ) <"$_lock"
1134 }
1135
1136 ctdb_reconfigure_release_lock ()
1137 {
1138     _ctdb_service_reconfigure_common
1139     _lock="${_d}/reconfigure_lock"
1140
1141     rm -f "$_lock"
1142 }
1143
1144 ctdb_replay_monitor_status ()
1145 {
1146     echo "Replaying previous status for this script due to reconfigure..."
1147     # Leading colon (':') is missing in some versions...
1148     _out=$(ctdb scriptstatus -Y | grep -E "^:?monitor:${script_name}:")
1149     # Output looks like this:
1150     # :monitor:60.nfs:1:ERROR:1314764004.030861:1314764004.035514:foo bar:
1151     # This is the cheapest way of getting fields in the middle.
1152     set -- $(IFS=":" ; echo $_out)
1153     _code="$3"
1154     _status="$4"
1155     # The error output field can include colons so we'll try to
1156     # preserve them.  The weak checking at the beginning tries to make
1157     # this work for both broken (no leading ':') and fixed output.
1158     _out="${_out%:}"
1159     _err_out="${_out#*monitor:${script_name}:*:*:*:*:}"
1160     case "$_status" in
1161         OK) : ;;  # Do nothing special.
1162         TIMEDOUT)
1163             # Recast this as an error, since we can't exit with the
1164             # correct negative number.
1165             _code=1
1166             _err_out="[Replay of TIMEDOUT scriptstatus - note incorrect return code.] ${_err_out}"
1167             ;;
1168         DISABLED)
1169             # Recast this as an OK, since we can't exit with the
1170             # correct negative number.
1171             _code=0
1172             _err_out="[Replay of DISABLED scriptstatus - note incorrect return code.] ${_err_out}"
1173             ;;
1174         *) : ;;  # Must be ERROR, do nothing special.
1175     esac
1176     if [ -n "$_err_out" ] ; then
1177         echo "$_err_out"
1178     fi
1179     exit $_code
1180 }
1181
1182 ctdb_service_check_reconfigure ()
1183 {
1184     assert_service_name
1185
1186     # We only care about some events in this function.  For others we
1187     # return now.
1188     case "$event_name" in
1189         monitor|ipreallocated|reconfigure) : ;;
1190         *) return 0 ;;
1191     esac
1192
1193     if ctdb_reconfigure_take_lock ; then
1194         # No events covered by this function are running, so proceed
1195         # with gay abandon.
1196         case "$event_name" in
1197             reconfigure)
1198                 (ctdb_service_reconfigure)
1199                 exit $?
1200                 ;;
1201             ipreallocated)
1202                 if ctdb_service_needs_reconfigure ; then
1203                     ctdb_service_reconfigure
1204                 fi
1205                 ;;
1206         esac
1207
1208         ctdb_reconfigure_release_lock
1209     else
1210         # Somebody else is running an event we don't want to collide
1211         # with.  We proceed with caution.
1212         case "$event_name" in
1213             reconfigure)
1214                 # Tell whoever called us to retry.
1215                 exit 2
1216                 ;;
1217             ipreallocated)
1218                 # Defer any scheduled reconfigure and just run the
1219                 # rest of the ipreallocated event, as per the
1220                 # eventscript.  There's an assumption here that the
1221                 # event doesn't depend on any scheduled reconfigure.
1222                 # This is true in the current code.
1223                 return 0
1224                 ;;
1225             monitor)
1226                 # There is most likely a reconfigure in progress so
1227                 # the service is possibly unstable.  As above, we
1228                 # defer any scheduled reconfigured.  We also replay
1229                 # the previous monitor status since that's the best
1230                 # information we have.
1231                 ctdb_replay_monitor_status
1232                 ;;
1233         esac
1234     fi
1235 }
1236
1237 ##################################################################
1238 # Does CTDB manage this service? - and associated auto-start/stop
1239
1240 ctdb_compat_managed_service ()
1241 {
1242     if [ "$1" = "yes" -a "$2" = "$service_name" ] ; then
1243         CTDB_MANAGED_SERVICES="$CTDB_MANAGED_SERVICES $2"
1244     fi
1245 }
1246
1247 is_ctdb_managed_service ()
1248 {
1249     assert_service_name
1250
1251     # $t is used just for readability and to allow better accurate
1252     # matching via leading/trailing spaces
1253     t=" $CTDB_MANAGED_SERVICES "
1254
1255     # Return 0 if "<space>$service_name<space>" appears in $t
1256     if [ "${t#* ${service_name} }" != "${t}" ] ; then
1257         return 0
1258     fi
1259
1260     # If above didn't match then update $CTDB_MANAGED_SERVICES for
1261     # backward compatibility and try again.
1262     ctdb_compat_managed_service "$CTDB_MANAGES_VSFTPD"   "vsftpd"
1263     ctdb_compat_managed_service "$CTDB_MANAGES_SAMBA"    "samba"
1264     ctdb_compat_managed_service "$CTDB_MANAGES_WINBIND"  "winbind"
1265     ctdb_compat_managed_service "$CTDB_MANAGES_HTTPD"    "apache2"
1266     ctdb_compat_managed_service "$CTDB_MANAGES_HTTPD"    "httpd"
1267     ctdb_compat_managed_service "$CTDB_MANAGES_ISCSI"    "iscsi"
1268     ctdb_compat_managed_service "$CTDB_MANAGES_CLAMD"    "clamd"
1269     ctdb_compat_managed_service "$CTDB_MANAGES_NFS"      "nfs"
1270     ctdb_compat_managed_service "$CTDB_MANAGES_NFS"      "nfs-ganesha-gpfs"
1271
1272     t=" $CTDB_MANAGED_SERVICES "
1273
1274     # Return 0 if "<space>$service_name<space>" appears in $t
1275     [ "${t#* ${service_name} }" != "${t}" ]
1276 }
1277
1278 ctdb_start_stop_service ()
1279 {
1280     assert_service_name
1281
1282     # Allow service-start/service-stop pseudo-events to start/stop
1283     # services when we're not auto-starting/stopping and we're not
1284     # monitoring.
1285     case "$event_name" in
1286         service-start)
1287             if is_ctdb_managed_service ; then
1288                 die 'service-start event not permitted when service is managed'
1289             fi
1290             if [ "$CTDB_SERVICE_AUTOSTARTSTOP" = "yes" ] ; then
1291                 die 'service-start event not permitted with $CTDB_SERVICE_AUTOSTARTSTOP = yes'
1292             fi
1293             ctdb_service_start
1294             exit $?
1295             ;;
1296         service-stop)
1297             if is_ctdb_managed_service ; then
1298                 die 'service-stop event not permitted when service is managed'
1299             fi
1300             if [ "$CTDB_SERVICE_AUTOSTARTSTOP" = "yes" ] ; then
1301                 die 'service-stop event not permitted with $CTDB_SERVICE_AUTOSTARTSTOP = yes'
1302             fi
1303             ctdb_service_stop
1304             exit $?
1305             ;;
1306     esac
1307
1308     # Do nothing unless configured to...
1309     [ "$CTDB_SERVICE_AUTOSTARTSTOP" = "yes" ] || return 0
1310
1311     [ "$event_name" = "monitor" ] || return 0
1312
1313     if is_ctdb_managed_service ; then
1314         if ! is_ctdb_previously_managed_service ; then
1315             echo "Starting service \"$service_name\" - now managed"
1316             background_with_logging ctdb_service_start
1317             exit $?
1318         fi
1319     else
1320         if is_ctdb_previously_managed_service ; then
1321             echo "Stopping service \"$service_name\" - no longer managed"
1322             background_with_logging ctdb_service_stop
1323             exit $?
1324         fi
1325     fi
1326 }
1327
1328 ctdb_service_start ()
1329 {
1330     # The service is marked managed if we've ever tried to start it.
1331     ctdb_service_managed
1332
1333     service_start || return $?
1334
1335     ctdb_counter_init
1336     ctdb_check_tcp_init
1337 }
1338
1339 ctdb_service_stop ()
1340 {
1341     ctdb_service_unmanaged
1342     service_stop
1343 }
1344
1345 # Default service_start() and service_stop() functions.
1346
1347 # These may be overridden in an eventscript.  When overriding, the
1348 # following convention must be followed.  If these functions are
1349 # called with no arguments then they may use internal logic to
1350 # determine whether the service is managed and, therefore, whether
1351 # they should take any action.  However, if the service name is
1352 # specified as an argument then an attempt must be made to start or
1353 # stop the service.  This is because the auto-start/stop code calls
1354 # them with the service name as an argument.
1355 service_start ()
1356 {
1357     service "$service_name" start
1358 }
1359
1360 service_stop ()
1361 {
1362     service "$service_name" stop
1363 }
1364
1365 ##################################################################
1366
1367 ctdb_standard_event_handler ()
1368 {
1369     case "$1" in
1370         status)
1371             ctdb_checkstatus
1372             exit
1373             ;;
1374         setstatus)
1375             shift
1376             ctdb_setstatus "$@"
1377             exit
1378             ;;
1379     esac
1380 }
1381
1382 # iptables doesn't like being re-entered, so flock-wrap it.
1383 iptables()
1384 {
1385         flock -w 30 $CTDB_VARDIR/iptables-ctdb.flock /sbin/iptables "$@"
1386 }
1387
1388 # AIX (and perhaps others?) doesn't have mktemp
1389 if ! which mktemp >/dev/null 2>&1 ; then
1390     mktemp ()
1391     {
1392         _dir=false
1393         if [ "$1" = "-d" ] ; then
1394             _dir=true
1395             shift
1396         fi
1397         _d="${TMPDIR:-/tmp}"
1398         _hex10=$(dd if=/dev/urandom count=20 2>/dev/null | \
1399             md5sum | \
1400             sed -e 's@\(..........\).*@\1@')
1401         _t="${_d}/tmp.${_hex10}"
1402         (
1403             umask 077
1404             if $_dir ; then
1405                 mkdir "$_t"
1406             else
1407                 >"$_t"
1408             fi
1409         )
1410         echo "$_t"
1411     }
1412 fi
1413
1414 ########################################################
1415 # tickle handling
1416 ########################################################
1417
1418 update_tickles ()
1419 {
1420         _port="$1"
1421
1422         tickledir="$CTDB_VARDIR/state/tickles"
1423         mkdir -p "$tickledir"
1424
1425         # Who am I?
1426         _pnn=$(ctdb pnn) ; _pnn=${_pnn#PNN:}
1427
1428         # What public IPs do I hold?
1429         _ips=$(ctdb -Y ip | awk -F: -v pnn=$_pnn '$3 == pnn {print $2}')
1430
1431         # IPs as a regexp choice
1432         _ipschoice="($(echo $_ips | sed -e 's/ /|/g' -e 's/\./\\\\./g'))"
1433
1434         # Record connections to our public IPs in a temporary file
1435         _my_connections="${tickledir}/${_port}.connections"
1436         rm -f "$_my_connections"
1437         netstat -tn |
1438         awk -v destpat="^${_ipschoice}:${_port}\$" \
1439           '$1 == "tcp" && $6 == "ESTABLISHED" && $4 ~ destpat {print $5, $4}' |
1440         sort >"$_my_connections"
1441
1442         # Record our current tickles in a temporary file
1443         _my_tickles="${tickledir}/${_port}.tickles"
1444         rm -f "$_my_tickles"
1445         for _i in $_ips ; do
1446                 ctdb -Y gettickles $_i $_port |
1447                 awk -F: 'NR > 1 { printf "%s:%s %s:%s\n", $2, $3, $4, $5 }'
1448         done |
1449         sort >"$_my_tickles"
1450
1451         # Add tickles for connections that we haven't already got tickles for
1452         comm -23 "$_my_connections" "$_my_tickles" |
1453         while read _src _dst ; do
1454                 ctdb addtickle $_src $_dst
1455         done
1456
1457         # Remove tickles for connections that are no longer there
1458         comm -13 "$_my_connections" "$_my_tickles" |
1459         while read _src _dst ; do
1460                 ctdb deltickle $_src $_dst
1461         done
1462
1463         rm -f "$_my_connections" "$_my_tickles"
1464 }
1465
1466 ########################################################
1467 # load a site local config file
1468 ########################################################
1469
1470 [ -n "$CTDB_RC_LOCAL" -a -x "$CTDB_RC_LOCAL" ] && {
1471         . "$CTDB_RC_LOCAL"
1472 }
1473
1474 [ -x $CTDB_BASE/rc.local ] && {
1475         . $CTDB_BASE/rc.local
1476 }
1477
1478 [ -d $CTDB_BASE/rc.local.d ] && {
1479         for i in $CTDB_BASE/rc.local.d/* ; do
1480                 [ -x "$i" ] && . "$i"
1481         done
1482 }
1483
1484 script_name="${0##*/}"       # basename
1485 service_fail_limit=1
1486 event_name="$1"