| blob | 49553c73559dc8bfc4c6218ca52d0b7ed43492db |
1 # Hey Emacs, this is a -*- shell-script -*- !!!
3 # utility functions for ctdb event scripts
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 }
16 #######################################
17 # pull in a system config file, if any
18 _loadconfig() {
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
28 if [ "$1" != "ctdb" ] ; then
29 loadconfig "ctdb"
30 fi
32 if [ -z "$1" ] ; then
33 return
34 fi
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
44 if [ "$1" = "ctdb" ] ; then
45 _config="${CTDB_BASE}/ctdbd.conf"
46 if [ -r "$_config" ] ; then
47 . "$_config"
48 fi
49 fi
50 }
52 loadconfig () {
53 _loadconfig "$@"
54 }
56 ##############################################################
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 }
74 die ()
75 {
76 _msg="$1"
77 _rc="${2:-1}"
79 echo "$_msg"
80 exit $_rc
81 }
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="$1" ; shift
89 case "$CTDB_LOGGING" in
90 file:*|"")
91 if [ -n "$CTDB_LOGGING" ] ; then
92 _file="${CTDB_LOGGING#file:}"
93 else
94 _file="/var/log/log.ctdb"
95 fi
96 {
97 if [ -n "$*" ] ; then
98 echo "$*"
99 else
100 cat
101 fi
102 } >>"$_file"
103 ;;
104 *)
105 # Handle all syslog:* variants here too. There's no tool to do
106 # the lossy things, so just use logger.
107 logger -t "ctdbd: ${_tag}" $*
108 ;;
109 esac
110 }
112 # When things are run in the background in an eventscript then logging
113 # output might get lost. This is the "solution". :-)
114 background_with_logging ()
115 {
116 (
117 "$@" 2>&1 </dev/null |
118 script_log "${script_name}&"
119 )&
121 return 0
122 }
124 ##############################################################
125 # check number of args for different events
126 ctdb_check_args ()
127 {
128 case "$1" in
129 takeip|releaseip)
130 if [ $# != 4 ]; then
131 echo "ERROR: must supply interface, IP and maskbits"
132 exit 1
133 fi
134 ;;
135 updateip)
136 if [ $# != 5 ]; then
137 echo "ERROR: must supply old interface, new interface, IP and maskbits"
138 exit 1
139 fi
140 ;;
141 esac
142 }
144 ##############################################################
145 # determine on what type of system (init style) we are running
146 detect_init_style()
147 {
148 # only do detection if not already set:
149 [ -z "$CTDB_INIT_STYLE" ] || return
151 if [ -x /sbin/startproc ]; then
152 CTDB_INIT_STYLE="suse"
153 elif [ -x /sbin/start-stop-daemon ]; then
154 CTDB_INIT_STYLE="debian"
155 else
156 CTDB_INIT_STYLE="redhat"
157 fi
158 }
160 ######################################################
161 # simulate /sbin/service on platforms that don't have it
162 # _service() makes it easier to hook the service() function for
163 # testing.
164 _service ()
165 {
166 _service_name="$1"
167 _op="$2"
169 # do nothing, when no service was specified
170 [ -z "$_service_name" ] && return
172 if [ -x /sbin/service ]; then
173 $_nice /sbin/service "$_service_name" "$_op"
174 elif [ -x $CTDB_ETCDIR/init.d/$_service_name ]; then
175 $_nice $CTDB_ETCDIR/init.d/$_service_name "$_op"
176 elif [ -x $CTDB_ETCDIR/rc.d/init.d/$_service_name ]; then
177 $_nice $CTDB_ETCDIR/rc.d/init.d/$_service_name "$_op"
178 fi
179 }
181 service()
182 {
183 _nice=""
184 _service "$@"
185 }
187 ######################################################
188 # simulate /sbin/service (niced) on platforms that don't have it
189 nice_service()
190 {
191 _nice="nice"
192 _service "$@"
193 }
195 ######################################################
196 # wrapper around /proc/ settings to allow them to be hooked
197 # for testing
198 # 1st arg is relative path under /proc/, 2nd arg is value to set
199 set_proc ()
200 {
201 echo "$2" >"/proc/$1"
202 }
204 ######################################################
205 # wrapper around getting file contents from /proc/ to allow
206 # this to be hooked for testing
207 # 1st arg is relative path under /proc/
208 get_proc ()
209 {
210 cat "/proc/$1"
211 }
213 ######################################################
214 # Print up to $_max kernel stack traces for processes named $_program
215 program_stack_traces ()
216 {
217 _prog="$1"
218 _max="${2:-1}"
220 _count=1
221 for _pid in $(pidof "$_prog") ; do
222 [ $_count -le $_max ] || break
224 # Do this first to avoid racing with process exit
225 _stack=$(get_proc "${_pid}/stack" 2>/dev/null)
226 if [ -n "$_stack" ] ; then
227 echo "Stack trace for ${_prog}[${_pid}]:"
228 echo "$_stack"
229 _count=$(($_count + 1))
230 fi
231 done
232 }
234 ######################################################
235 # Check that an RPC service is healthy -
236 # this includes allowing a certain number of failures
237 # before marking the NFS service unhealthy.
238 #
239 # usage: nfs_check_rpc_service SERVICE_NAME [ triple ...]
240 #
241 # each triple is a set of 3 arguments: an operator, a
242 # fail count limit and an action string.
243 #
244 # For example:
245 #
246 # nfs_check_rpc_service "lockd" \
247 # -ge 15 "verbose restart unhealthy" \
248 # -eq 10 "restart:bs"
249 #
250 # says that if lockd is down for 15 iterations then do
251 # a verbose restart of lockd and mark the node unhealthy.
252 # Before this, after 10 iterations of failure, the
253 # service is restarted silently in the background.
254 # Order is important: the number of failures need to be
255 # specified in reverse order because processing stops
256 # after the first condition that is true.
257 ######################################################
258 nfs_check_rpc_service ()
259 {
260 _prog_name="$1" ; shift
262 if _nfs_check_rpc_common "$_prog_name" ; then
263 return
264 fi
266 while [ -n "$3" ] ; do
267 if _nfs_check_rpc_action "$1" "$2" "$3" ; then
268 break
269 fi
270 shift 3
271 done
272 }
274 # The new way of doing things...
275 nfs_check_rpc_services ()
276 {
277 # Files must end with .check - avoids editor backups, RPM fu, ...
278 for _f in "${CTDB_BASE}/nfs-rpc-checks.d/"[0-9][0-9].*.check ; do
279 _t="${_f%.check}"
280 _prog_name="${_t##*/[0-9][0-9].}"
282 if _nfs_check_rpc_common "$_prog_name" ; then
283 # This RPC service is up, check next service...
284 continue
285 fi
287 # Check each line in the file in turn until one of the limit
288 # checks is hit...
289 while read _cmp _lim _rest ; do
290 # Skip comments
291 case "$_cmp" in
292 \#*) continue ;;
293 esac
295 if _nfs_check_rpc_action "$_cmp" "$_lim" "$_rest" ; then
296 # Limit was hit on this line, no further checking...
297 break
298 fi
299 done <"$_f"
300 done
301 }
303 _nfs_check_rpc_common ()
304 {
305 _prog_name="$1"
307 # Some platforms don't have separate programs for all services.
308 case "$_prog_name" in
309 statd)
310 which "rpc.${_prog_name}" >/dev/null 2>&1 || return 0
311 esac
313 case "$_prog_name" in
314 nfsd)
315 _rpc_prog=nfs
316 _version=3
317 ;;
318 mountd)
319 _rpc_prog=mountd
320 _version=1
321 ;;
322 rquotad)
323 _rpc_prog=rquotad
324 _version=1
325 ;;
326 lockd)
327 _rpc_prog=nlockmgr
328 _version=4
329 ;;
330 statd)
331 _rpc_prog=status
332 _version=1
333 ;;
334 *)
335 echo "Internal error: unknown RPC program \"$_prog_name\"."
336 exit 1
337 esac
339 _service_name="nfs_${_prog_name}"
341 if ctdb_check_rpc "$_rpc_prog" $_version >/dev/null ; then
342 ctdb_counter_init "$_service_name"
343 return 0
344 fi
346 ctdb_counter_incr "$_service_name"
348 return 1
349 }
351 _nfs_check_rpc_action ()
352 {
353 _cmp="$1"
354 _limit="$2"
355 _actions="$3"
357 if ctdb_check_counter "quiet" "$_cmp" "$_limit" "$_service_name" ; then
358 return 1
359 fi
361 for _action in $_actions ; do
362 case "$_action" in
363 verbose)
364 echo "$ctdb_check_rpc_out"
365 ;;
366 restart)
367 _nfs_restart_rpc_service "$_prog_name"
368 ;;
369 restart:b)
370 _nfs_restart_rpc_service "$_prog_name" true
371 ;;
372 unhealthy)
373 exit 1
374 ;;
375 *)
376 echo "Internal error: unknown action \"$_action\"."
377 exit 1
378 esac
379 done
381 return 0
382 }
384 _nfs_restart_rpc_service ()
385 {
386 _prog_name="$1"
387 _background="${2:-false}"
389 if $_background ; then
390 _maybe_background="background_with_logging"
391 else
392 _maybe_background=""
393 fi
395 _p="rpc.${_prog_name}"
397 case "$_prog_name" in
398 nfsd)
399 echo "Trying to restart NFS service"
400 $_maybe_background startstop_nfs restart
401 ;;
402 mountd)
403 echo "Trying to restart $_prog_name [${_p}]"
404 killall -q -9 "$_p"
405 $_maybe_background $_p ${MOUNTD_PORT:+-p} $MOUNTD_PORT
406 ;;
407 rquotad)
408 echo "Trying to restart $_prog_name [${_p}]"
409 killall -q -9 "$_p"
410 $_maybe_background $_p ${RQUOTAD_PORT:+-p} $RQUOTAD_PORT
411 ;;
412 lockd)
413 echo "Trying to restart lock manager service"
414 $_maybe_background startstop_nfslock restart
415 ;;
416 statd)
417 echo "Trying to restart $_prog_name [${_p}]"
418 killall -q -9 "$_p"
419 $_maybe_background $_p \
420 ${STATD_HOSTNAME:+-n} $STATD_HOSTNAME \
421 ${STATD_PORT:+-p} $STATD_PORT \
422 ${STATD_OUTGOING_PORT:+-o} $STATD_OUTGOING_PORT
423 ;;
424 *)
425 echo "Internal error: unknown RPC program \"$_prog_name\"."
426 exit 1
427 esac
428 }
430 ######################################################
431 # check that a rpc server is registered with portmap
432 # and responding to requests
433 # usage: ctdb_check_rpc SERVICE_NAME VERSION
434 ######################################################
435 ctdb_check_rpc ()
436 {
437 progname="$1"
438 version="$2"
440 _localhost="${CTDB_RPCINFO_LOCALHOST:-127.0.0.1}"
442 if ! ctdb_check_rpc_out=$(rpcinfo -u $_localhost $progname $version 2>&1) ; then
443 ctdb_check_rpc_out="ERROR: $progname failed RPC check:
444 $ctdb_check_rpc_out"
445 echo "$ctdb_check_rpc_out"
446 return 1
447 fi
448 }
450 ######################################################
451 # Ensure $service_name is set
452 assert_service_name ()
453 {
454 [ -n "$service_name" ] || die "INTERNAL ERROR: \$service_name not set"
455 }
457 ######################################################
458 # check a set of directories is available
459 # return 1 on a missing directory
460 # directories are read from stdin
461 ######################################################
462 ctdb_check_directories_probe()
463 {
464 while IFS="" read d ; do
465 case "$d" in
466 *%*)
467 continue
468 ;;
469 *)
470 [ -d "${d}/." ] || return 1
471 esac
472 done
473 }
475 ######################################################
476 # check a set of directories is available
477 # directories are read from stdin
478 ######################################################
479 ctdb_check_directories()
480 {
481 ctdb_check_directories_probe || {
482 echo "ERROR: $service_name directory \"$d\" not available"
483 exit 1
484 }
485 }
487 ######################################################
488 # check a set of tcp ports
489 # usage: ctdb_check_tcp_ports <ports...>
490 ######################################################
492 # This flag file is created when a service is initially started. It
493 # is deleted the first time TCP port checks for that service succeed.
494 # Until then ctdb_check_tcp_ports() prints a more subtle "error"
495 # message if a port check fails.
496 _ctdb_check_tcp_common ()
497 {
498 assert_service_name
499 _ctdb_service_started_file="$ctdb_fail_dir/$service_name.started"
500 }
502 ctdb_check_tcp_init ()
503 {
504 _ctdb_check_tcp_common
505 mkdir -p "${_ctdb_service_started_file%/*}" # dirname
506 touch "$_ctdb_service_started_file"
507 }
509 # Check whether something is listening on all of the given TCP ports
510 # using the "ctdb checktcpport" command.
511 ctdb_check_tcp_ports()
512 {
513 if [ -z "$1" ] ; then
514 echo "INTERNAL ERROR: ctdb_check_tcp_ports - no ports specified"
515 exit 1
516 fi
518 for _p ; do # process each function argument (port)
519 _cmd="ctdb checktcpport $_p"
520 _out=$($_cmd 2>&1)
521 _ret=$?
522 case "$_ret" in
523 0)
524 _ctdb_check_tcp_common
525 if [ ! -f "$_ctdb_service_started_file" ] ; then
526 echo "ERROR: $service_name tcp port $_p is not responding"
527 debug "\"ctdb checktcpport $_p\" was able to bind to port"
528 else
529 echo "INFO: $service_name tcp port $_p is not responding"
530 fi
532 return 1
533 ;;
534 98)
535 # Couldn't bind, something already listening, next port...
536 continue
537 ;;
538 *)
539 echo "ERROR: unexpected error running \"ctdb checktcpport\""
540 debug <<EOF
541 ctdb checktcpport (exited with $_ret) with output:
542 $_out"
543 EOF
544 return $_ret
545 esac
546 done
548 # All ports listening
549 _ctdb_check_tcp_common
550 rm -f "$_ctdb_service_started_file"
551 return 0
552 }
554 ######################################################
555 # check a unix socket
556 # usage: ctdb_check_unix_socket SERVICE_NAME <socket_path>
557 ######################################################
558 ctdb_check_unix_socket() {
559 socket_path="$1"
560 [ -z "$socket_path" ] && return
562 if ! netstat --unix -a -n | grep -q "^unix.*LISTEN.*${socket_path}$"; then
563 echo "ERROR: $service_name socket $socket_path not found"
564 return 1
565 fi
566 }
568 ######################################################
569 # check a command returns zero status
570 # usage: ctdb_check_command <command>
571 ######################################################
572 ctdb_check_command ()
573 {
574 _out=$("$@" 2>&1) || {
575 echo "ERROR: $* returned error"
576 echo "$_out" | debug
577 exit 1
578 }
579 }
581 ################################################
582 # kill off any TCP connections with the given IP
583 ################################################
584 kill_tcp_connections ()
585 {
586 _ip="$1"
588 _oneway=false
589 if [ "$2" = "oneway" ] ; then
590 _oneway=true
591 fi
593 get_tcp_connections_for_ip "$_ip" | {
594 _killcount=0
595 _connections=""
596 _nl="
597 "
598 while read _dst _src; do
599 _destport="${_dst##*:}"
600 __oneway=$_oneway
601 case $_destport in
602 # we only do one-way killtcp for CIFS
603 139|445) __oneway=true ;;
604 esac
606 echo "Killing TCP connection $_src $_dst"
607 _connections="${_connections}${_nl}${_src} ${_dst}"
608 if ! $__oneway ; then
609 _connections="${_connections}${_nl}${_dst} ${_src}"
610 fi
612 _killcount=$(($_killcount + 1))
613 done
615 if [ $_killcount -eq 0 ] ; then
616 return
617 fi
619 echo "$_connections" | ctdb killtcp || {
620 echo "Failed to send killtcp control"
621 return
622 }
624 _count=0
625 while : ; do
626 _remaining=$(get_tcp_connections_for_ip $_ip | wc -l)
628 if [ $_remaining -eq 0 ] ; then
629 echo "Killed $_killcount TCP connections to released IP $_ip"
630 return
631 fi
633 _count=$(($_count + 1))
634 if [ $_count -gt 3 ] ; then
635 echo "Timed out killing tcp connections for IP $_ip ($_remaining remaining)"
636 return
637 fi
639 echo "Waiting for $_remaining connections to be killed for IP $_ip"
640 sleep 1
641 done
642 }
643 }
645 ##################################################################
646 # kill off the local end for any TCP connections with the given IP
647 ##################################################################
648 kill_tcp_connections_local_only ()
649 {
650 kill_tcp_connections "$1" "oneway"
651 }
653 ##################################################################
654 # tickle any TCP connections with the given IP
655 ##################################################################
656 tickle_tcp_connections ()
657 {
658 _ip="$1"
660 get_tcp_connections_for_ip "$_ip" |
661 {
662 _failed=false
664 while read dest src; do
665 echo "Tickle TCP connection $src $dest"
666 ctdb tickle $src $dest >/dev/null 2>&1 || _failed=true
667 echo "Tickle TCP connection $dest $src"
668 ctdb tickle $dest $src >/dev/null 2>&1 || _failed=true
669 done
671 if $_failed ; then
672 echo "Failed to send tickle control"
673 fi
674 }
675 }
677 get_tcp_connections_for_ip ()
678 {
679 _ip="$1"
681 netstat -tn | awk -v ip=$_ip \
682 'index($1, "tcp") == 1 && \
683 (index($4, ip ":") == 1 || index($4, "::ffff:" ip ":") == 1) \
684 && $6 == "ESTABLISHED" \
685 {print $4" "$5}'
686 }
688 ########################################################
689 # start/stop the Ganesha nfs service
690 ########################################################
691 startstop_ganesha()
692 {
693 _service_name="nfs-ganesha-$CTDB_CLUSTER_FILESYSTEM_TYPE"
694 case "$1" in
695 start)
696 service "$_service_name" start
697 ;;
698 stop)
699 service "$_service_name" stop
700 ;;
701 restart)
702 service "$_service_name" restart
703 ;;
704 esac
705 }
707 ########################################################
708 # start/stop the nfs service on different platforms
709 ########################################################
710 startstop_nfs() {
711 PLATFORM="unknown"
712 [ -x $CTDB_ETCDIR/init.d/nfsserver ] && {
713 PLATFORM="sles"
714 }
715 [ -x $CTDB_ETCDIR/init.d/nfslock -o \
716 -r /usr/lib/systemd/system/nfs-lock.service ] && {
717 PLATFORM="rhel"
718 }
720 case $PLATFORM in
721 sles)
722 case $1 in
723 start)
724 service nfsserver start
725 ;;
726 stop)
727 service nfsserver stop > /dev/null 2>&1
728 ;;
729 restart)
730 set_proc "fs/nfsd/threads" 0
731 service nfsserver stop > /dev/null 2>&1
732 pkill -9 nfsd
733 nfs_dump_some_threads
734 service nfsserver start
735 ;;
736 esac
737 ;;
738 rhel)
739 case $1 in
740 start)
741 service nfslock start
742 service nfs start
743 ;;
744 stop)
745 service nfs stop
746 service nfslock stop
747 ;;
748 restart)
749 set_proc "fs/nfsd/threads" 0
750 service nfs stop > /dev/null 2>&1
751 service nfslock stop > /dev/null 2>&1
752 pkill -9 nfsd
753 nfs_dump_some_threads
754 service nfslock start
755 service nfs start
756 ;;
757 esac
758 ;;
759 *)
760 echo "Unknown platform. NFS is not supported with ctdb"
761 exit 1
762 ;;
763 esac
764 }
766 # Dump up to the configured number of nfsd thread backtraces.
767 nfs_dump_some_threads ()
768 {
769 _num="${CTDB_NFS_DUMP_STUCK_THREADS:-5}"
770 [ $_num -gt 0 ] || return 0
772 program_stack_traces "nfsd" $_num
773 }
775 ########################################################
776 # start/stop the nfs lockmanager service on different platforms
777 ########################################################
778 startstop_nfslock() {
779 PLATFORM="unknown"
780 [ -x $CTDB_ETCDIR/init.d/nfsserver ] && {
781 PLATFORM="sles"
782 }
783 [ -x $CTDB_ETCDIR/init.d/nfslock -o \
784 -r /usr/lib/systemd/system/nfs-lock.service ] && {
785 PLATFORM="rhel"
786 }
788 case $PLATFORM in
789 sles)
790 # for sles there is no service for lockmanager
791 # so we instead just shutdown/restart nfs
792 case $1 in
793 start)
794 service nfsserver start
795 ;;
796 stop)
797 service nfsserver stop > /dev/null 2>&1
798 ;;
799 restart)
800 service nfsserver stop > /dev/null 2>&1
801 service nfsserver start
802 ;;
803 esac
804 ;;
805 rhel)
806 case $1 in
807 start)
808 service nfslock start
809 ;;
810 stop)
811 service nfslock stop > /dev/null 2>&1
812 ;;
813 restart)
814 service nfslock stop > /dev/null 2>&1
815 service nfslock start
816 ;;
817 esac
818 ;;
819 *)
820 echo "Unknown platform. NFS locking is not supported with ctdb"
821 exit 1
822 ;;
823 esac
824 }
826 # Periodically update the statd database
827 nfs_statd_update ()
828 {
829 _update_period="$1"
831 _statd_update_trigger="$service_state_dir/update-trigger"
832 [ -f "$_statd_update_trigger" ] || touch "$_statd_update_trigger"
834 _last_update=$(stat --printf="%Y" "$_statd_update_trigger")
835 _current_time=$(date +"%s")
836 if [ $(( $_current_time - $_last_update)) -ge $_update_period ] ; then
837 touch "$_statd_update_trigger"
838 $CTDB_BASE/statd-callout updatelocal &
839 $CTDB_BASE/statd-callout updateremote &
840 fi
841 }
843 ########################################################
845 add_ip_to_iface ()
846 {
847 _iface=$1
848 _ip=$2
849 _maskbits=$3
851 # Ensure interface is up
852 ip link set "$_iface" up || \
853 die "Failed to bringup interface $_iface"
855 ip addr add "$_ip/$_maskbits" brd + dev "$_iface" || {
856 echo "Failed to add $_ip/$_maskbits on dev $_iface"
857 return 1
858 }
859 }
861 delete_ip_from_iface()
862 {
863 _iface=$1
864 _ip=$2
865 _maskbits=$3
867 # This could be set globally for all interfaces but it is probably
868 # better to avoid surprises, so limit it the interfaces where CTDB
869 # has public IP addresses. There isn't anywhere else convenient
870 # to do this so just set it each time. This is much cheaper than
871 # remembering and re-adding secondaries.
872 set_proc "sys/net/ipv4/conf/${_iface}/promote_secondaries" 1
874 ip addr del "$_ip/$_maskbits" dev "$_iface" || {
875 echo "Failed to del $_ip on dev $_iface"
876 return 1
877 }
878 }
880 # If the given IP is hosted then print 2 items: maskbits and iface
881 ip_maskbits_iface ()
882 {
883 _addr="$1"
885 ip addr show to "${_addr}/32" 2>/dev/null | \
886 awk '$1 == "inet" { print gensub(".*/", "", 1, $2), $NF }'
887 }
889 drop_ip ()
890 {
891 _addr="${1%/*}" # Remove optional maskbits
893 set -- $(ip_maskbits_iface $_addr)
894 if [ -n "$1" ] ; then
895 _maskbits="$1"
896 _iface="$2"
897 echo "Removing public address $_addr/$_maskbits from device $_iface"
898 delete_ip_from_iface $_iface $_addr $_maskbits >/dev/null 2>&1
899 fi
900 }
902 drop_all_public_ips ()
903 {
904 while read _ip _x ; do
905 drop_ip "$_ip"
906 done <"${CTDB_PUBLIC_ADDRESSES:-/dev/null}"
907 }
909 ########################################################
910 # Simple counters
911 _ctdb_counter_common () {
912 _service_name="${1:-${service_name:-${script_name}}}"
913 _counter_file="$ctdb_fail_dir/$_service_name"
914 mkdir -p "${_counter_file%/*}" # dirname
915 }
916 ctdb_counter_init () {
917 _ctdb_counter_common "$1"
919 >"$_counter_file"
920 }
921 ctdb_counter_incr () {
922 _ctdb_counter_common "$1"
924 # unary counting!
925 echo -n 1 >> "$_counter_file"
926 }
927 ctdb_check_counter () {
928 _msg="${1:-error}" # "error" - anything else is silent on fail
929 _op="${2:--ge}" # an integer operator supported by test
930 _limit="${3:-${service_fail_limit}}"
931 shift 3
932 _ctdb_counter_common "$1"
934 # unary counting!
935 _size=$(stat -c "%s" "$_counter_file" 2>/dev/null || echo 0)
936 _hit=false
937 if [ "$_op" != "%" ] ; then
938 if [ $_size $_op $_limit ] ; then
939 _hit=true
940 fi
941 else
942 if [ $(($_size $_op $_limit)) -eq 0 ] ; then
943 _hit=true
944 fi
945 fi
946 if $_hit ; then
947 if [ "$_msg" = "error" ] ; then
948 echo "ERROR: $_size consecutive failures for $_service_name, marking node unhealthy"
949 exit 1
950 else
951 return 1
952 fi
953 fi
954 }
956 ########################################################
958 ctdb_status_dir="$CTDB_VARDIR/state/service_status"
959 ctdb_fail_dir="$CTDB_VARDIR/state/failcount"
961 ctdb_setup_service_state_dir ()
962 {
963 service_state_dir="$CTDB_VARDIR/state/service_state/${1:-${service_name}}"
964 mkdir -p "$service_state_dir" || {
965 echo "Error creating state dir \"$service_state_dir\""
966 exit 1
967 }
968 }
970 ########################################################
971 # Managed status history, for auto-start/stop
973 ctdb_managed_dir="$CTDB_VARDIR/state/managed_history"
975 _ctdb_managed_common ()
976 {
977 _ctdb_managed_file="$ctdb_managed_dir/$service_name"
978 }
980 ctdb_service_managed ()
981 {
982 _ctdb_managed_common
983 mkdir -p "$ctdb_managed_dir"
984 touch "$_ctdb_managed_file"
985 }
987 ctdb_service_unmanaged ()
988 {
989 _ctdb_managed_common
990 rm -f "$_ctdb_managed_file"
991 }
993 is_ctdb_previously_managed_service ()
994 {
995 _ctdb_managed_common
996 [ -f "$_ctdb_managed_file" ]
997 }
999 ########################################################
1000 # Check and set status
1002 log_status_cat ()
1003 {
1004 echo "node is \"$1\", \"${script_name}\" reports problem: $(cat $2)"
1005 }
1007 ctdb_checkstatus ()
1008 {
1009 if [ -r "$ctdb_status_dir/$script_name/unhealthy" ] ; then
1010 log_status_cat "unhealthy" "$ctdb_status_dir/$script_name/unhealthy"
1011 return 1
1012 elif [ -r "$ctdb_status_dir/$script_name/banned" ] ; then
1013 log_status_cat "banned" "$ctdb_status_dir/$script_name/banned"
1014 return 2
1015 else
1016 return 0
1017 fi
1018 }
1020 ctdb_setstatus ()
1021 {
1022 d="$ctdb_status_dir/$script_name"
1023 case "$1" in
1024 unhealthy|banned)
1025 mkdir -p "$d"
1026 cat "$2" >"$d/$1"
1027 ;;
1028 *)
1029 for i in "banned" "unhealthy" ; do
1030 rm -f "$d/$i"
1031 done
1032 ;;
1033 esac
1034 }
1036 ##################################################################
1037 # Reconfigure a service on demand
1039 _ctdb_service_reconfigure_common ()
1040 {
1041 _d="$ctdb_status_dir/${service_name}"
1042 mkdir -p "$_d"
1043 _ctdb_service_reconfigure_flag="$_d/reconfigure"
1044 }
1046 ctdb_service_needs_reconfigure ()
1047 {
1048 _ctdb_service_reconfigure_common
1049 [ -e "$_ctdb_service_reconfigure_flag" ]
1050 }
1052 ctdb_service_set_reconfigure ()
1053 {
1054 _ctdb_service_reconfigure_common
1055 >"$_ctdb_service_reconfigure_flag"
1056 }
1058 ctdb_service_unset_reconfigure ()
1059 {
1060 _ctdb_service_reconfigure_common
1061 rm -f "$_ctdb_service_reconfigure_flag"
1062 }
1064 ctdb_service_reconfigure ()
1065 {
1066 echo "Reconfiguring service \"${service_name}\"..."
1067 ctdb_service_unset_reconfigure
1068 service_reconfigure || return $?
1069 ctdb_counter_init
1070 }
1072 # Default service_reconfigure() function does nothing.
1073 service_reconfigure ()
1074 {
1075 :
1076 }
1078 ctdb_reconfigure_take_lock ()
1079 {
1080 _ctdb_service_reconfigure_common
1081 _lock="${_d}/reconfigure_lock"
1082 mkdir -p "${_lock%/*}" # dirname
1083 touch "$_lock"
1085 (
1086 flock 0
1087 # This is overkill but will work if we need to extend this to
1088 # allow certain events to run multiple times in parallel
1089 # (e.g. takeip) and write multiple PIDs to the file.
1090 read _locker_event
1091 if [ -n "$_locker_event" ] ; then
1092 while read _pid ; do
1093 if [ -n "$_pid" -a "$_pid" != $$ ] && \
1094 kill -0 "$_pid" 2>/dev/null ; then
1095 exit 1
1096 fi
1097 done
1098 fi
1100 printf "%s\n%s\n" "$event_name" $$ >"$_lock"
1101 exit 0
1102 ) <"$_lock"
1103 }
1105 ctdb_reconfigure_release_lock ()
1106 {
1107 _ctdb_service_reconfigure_common
1108 _lock="${_d}/reconfigure_lock"
1110 rm -f "$_lock"
1111 }
1113 ctdb_replay_monitor_status ()
1114 {
1115 echo "Replaying previous status for this script due to reconfigure..."
1116 # Leading colon (':') is missing in some versions...
1117 _out=$(ctdb scriptstatus -Y | grep -E "^:?monitor:${script_name}:")
1118 # Output looks like this:
1119 # :monitor:60.nfs:1:ERROR:1314764004.030861:1314764004.035514:foo bar:
1120 # This is the cheapest way of getting fields in the middle.
1121 set -- $(IFS=":" ; echo $_out)
1122 _code="$3"
1123 _status="$4"
1124 # The error output field can include colons so we'll try to
1125 # preserve them. The weak checking at the beginning tries to make
1126 # this work for both broken (no leading ':') and fixed output.
1127 _out="${_out%:}"
1128 _err_out="${_out#*monitor:${script_name}:*:*:*:*:}"
1129 case "$_status" in
1130 OK) : ;; # Do nothing special.
1131 TIMEDOUT)
1132 # Recast this as an error, since we can't exit with the
1133 # correct negative number.
1134 _code=1
1135 _err_out="[Replay of TIMEDOUT scriptstatus - note incorrect return code.] ${_err_out}"
1136 ;;
1137 DISABLED)
1138 # Recast this as an OK, since we can't exit with the
1139 # correct negative number.
1140 _code=0
1141 _err_out="[Replay of DISABLED scriptstatus - note incorrect return code.] ${_err_out}"
1142 ;;
1143 *) : ;; # Must be ERROR, do nothing special.
1144 esac
1145 if [ -n "$_err_out" ] ; then
1146 echo "$_err_out"
1147 fi
1148 exit $_code
1149 }
1151 ctdb_service_check_reconfigure ()
1152 {
1153 assert_service_name
1155 # We only care about some events in this function. For others we
1156 # return now.
1157 case "$event_name" in
1158 monitor|ipreallocated|reconfigure) : ;;
1159 *) return 0 ;;
1160 esac
1162 if ctdb_reconfigure_take_lock ; then
1163 # No events covered by this function are running, so proceed
1164 # with gay abandon.
1165 case "$event_name" in
1166 reconfigure)
1167 (ctdb_service_reconfigure)
1168 exit $?
1169 ;;
1170 ipreallocated)
1171 if ctdb_service_needs_reconfigure ; then
1172 ctdb_service_reconfigure
1173 fi
1174 ;;
1175 esac
1177 ctdb_reconfigure_release_lock
1178 else
1179 # Somebody else is running an event we don't want to collide
1180 # with. We proceed with caution.
1181 case "$event_name" in
1182 reconfigure)
1183 # Tell whoever called us to retry.
1184 exit 2
1185 ;;
1186 ipreallocated)
1187 # Defer any scheduled reconfigure and just run the
1188 # rest of the ipreallocated event, as per the
1189 # eventscript. There's an assumption here that the
1190 # event doesn't depend on any scheduled reconfigure.
1191 # This is true in the current code.
1192 return 0
1193 ;;
1194 monitor)
1195 # There is most likely a reconfigure in progress so
1196 # the service is possibly unstable. As above, we
1197 # defer any scheduled reconfigured. We also replay
1198 # the previous monitor status since that's the best
1199 # information we have.
1200 ctdb_replay_monitor_status
1201 ;;
1202 esac
1203 fi
1204 }
1206 ##################################################################
1207 # Does CTDB manage this service? - and associated auto-start/stop
1209 ctdb_compat_managed_service ()
1210 {
1211 if [ "$1" = "yes" -a "$2" = "$service_name" ] ; then
1212 CTDB_MANAGED_SERVICES="$CTDB_MANAGED_SERVICES $2"
1213 fi
1214 }
1216 is_ctdb_managed_service ()
1217 {
1218 assert_service_name
1220 # $t is used just for readability and to allow better accurate
1221 # matching via leading/trailing spaces
1222 t=" $CTDB_MANAGED_SERVICES "
1224 # Return 0 if "<space>$service_name<space>" appears in $t
1225 if [ "${t#* ${service_name} }" != "${t}" ] ; then
1226 return 0
1227 fi
1229 # If above didn't match then update $CTDB_MANAGED_SERVICES for
1230 # backward compatibility and try again.
1231 ctdb_compat_managed_service "$CTDB_MANAGES_VSFTPD" "vsftpd"
1232 ctdb_compat_managed_service "$CTDB_MANAGES_SAMBA" "samba"
1233 ctdb_compat_managed_service "$CTDB_MANAGES_WINBIND" "winbind"
1234 ctdb_compat_managed_service "$CTDB_MANAGES_HTTPD" "apache2"
1235 ctdb_compat_managed_service "$CTDB_MANAGES_HTTPD" "httpd"
1236 ctdb_compat_managed_service "$CTDB_MANAGES_ISCSI" "iscsi"
1237 ctdb_compat_managed_service "$CTDB_MANAGES_CLAMD" "clamd"
1238 ctdb_compat_managed_service "$CTDB_MANAGES_NFS" "nfs"
1239 ctdb_compat_managed_service "$CTDB_MANAGES_NFS" "nfs-ganesha-gpfs"
1241 t=" $CTDB_MANAGED_SERVICES "
1243 # Return 0 if "<space>$service_name<space>" appears in $t
1244 [ "${t#* ${service_name} }" != "${t}" ]
1245 }
1247 ctdb_start_stop_service ()
1248 {
1249 assert_service_name
1251 # Allow service-start/service-stop pseudo-events to start/stop
1252 # services when we're not auto-starting/stopping and we're not
1253 # monitoring.
1254 case "$event_name" in
1255 service-start)
1256 if is_ctdb_managed_service ; then
1257 die 'service-start event not permitted when service is managed'
1258 fi
1259 if [ "$CTDB_SERVICE_AUTOSTARTSTOP" = "yes" ] ; then
1260 die 'service-start event not permitted with $CTDB_SERVICE_AUTOSTARTSTOP = yes'
1261 fi
1262 ctdb_service_start
1263 exit $?
1264 ;;
1265 service-stop)
1266 if is_ctdb_managed_service ; then
1267 die 'service-stop event not permitted when service is managed'
1268 fi
1269 if [ "$CTDB_SERVICE_AUTOSTARTSTOP" = "yes" ] ; then
1270 die 'service-stop event not permitted with $CTDB_SERVICE_AUTOSTARTSTOP = yes'
1271 fi
1272 ctdb_service_stop
1273 exit $?
1274 ;;
1275 esac
1277 # Do nothing unless configured to...
1278 [ "$CTDB_SERVICE_AUTOSTARTSTOP" = "yes" ] || return 0
1280 [ "$event_name" = "monitor" ] || return 0
1282 if is_ctdb_managed_service ; then
1283 if ! is_ctdb_previously_managed_service ; then
1284 echo "Starting service \"$service_name\" - now managed"
1285 background_with_logging ctdb_service_start
1286 exit $?
1287 fi
1288 else
1289 if is_ctdb_previously_managed_service ; then
1290 echo "Stopping service \"$service_name\" - no longer managed"
1291 background_with_logging ctdb_service_stop
1292 exit $?
1293 fi
1294 fi
1295 }
1297 ctdb_service_start ()
1298 {
1299 # The service is marked managed if we've ever tried to start it.
1300 ctdb_service_managed
1302 service_start || return $?
1304 ctdb_counter_init
1305 ctdb_check_tcp_init
1306 }
1308 ctdb_service_stop ()
1309 {
1310 ctdb_service_unmanaged
1311 service_stop
1312 }
1314 # Default service_start() and service_stop() functions.
1316 # These may be overridden in an eventscript.
1317 service_start ()
1318 {
1319 service "$service_name" start
1320 }
1322 service_stop ()
1323 {
1324 service "$service_name" stop
1325 }
1327 ##################################################################
1329 ctdb_standard_event_handler ()
1330 {
1331 case "$1" in
1332 status)
1333 ctdb_checkstatus
1334 exit
1335 ;;
1336 setstatus)
1337 shift
1338 ctdb_setstatus "$@"
1339 exit
1340 ;;
1341 esac
1342 }
1344 # iptables doesn't like being re-entered, so flock-wrap it.
1345 iptables()
1346 {
1347 flock -w 30 $CTDB_VARDIR/iptables-ctdb.flock /sbin/iptables "$@"
1348 }
1350 # AIX (and perhaps others?) doesn't have mktemp
1351 if ! which mktemp >/dev/null 2>&1 ; then
1352 mktemp ()
1353 {
1354 _dir=false
1355 if [ "$1" = "-d" ] ; then
1356 _dir=true
1357 shift
1358 fi
1359 _d="${TMPDIR:-/tmp}"
1360 _hex10=$(dd if=/dev/urandom count=20 2>/dev/null | \
1361 md5sum | \
1362 sed -e 's@\(..........\).*@\1@')
1363 _t="${_d}/tmp.${_hex10}"
1364 (
1365 umask 077
1366 if $_dir ; then
1367 mkdir "$_t"
1368 else
1369 >"$_t"
1370 fi
1371 )
1372 echo "$_t"
1373 }
1374 fi
1376 ########################################################
1377 # tickle handling
1378 ########################################################
1380 update_tickles ()
1381 {
1382 _port="$1"
1384 tickledir="$CTDB_VARDIR/state/tickles"
1385 mkdir -p "$tickledir"
1387 # Who am I?
1388 _pnn=$(ctdb pnn) ; _pnn=${_pnn#PNN:}
1390 # What public IPs do I hold?
1391 _ips=$(ctdb -Y ip | awk -F: -v pnn=$_pnn '$3 == pnn {print $2}')
1393 # IPs as a regexp choice
1394 _ipschoice="($(echo $_ips | sed -e 's/ /|/g' -e 's/\./\\\\./g'))"
1396 # Record connections to our public IPs in a temporary file
1397 _my_connections="${tickledir}/${_port}.connections"
1398 rm -f "$_my_connections"
1399 netstat -tn |
1400 awk -v destpat="^${_ipschoice}:${_port}\$" \
1401 '$1 == "tcp" && $6 == "ESTABLISHED" && $4 ~ destpat {print $5, $4}' |
1402 sort >"$_my_connections"
1404 # Record our current tickles in a temporary file
1405 _my_tickles="${tickledir}/${_port}.tickles"
1406 rm -f "$_my_tickles"
1407 for _i in $_ips ; do
1408 ctdb -Y gettickles $_i $_port |
1409 awk -F: 'NR > 1 { printf "%s:%s %s:%s\n", $2, $3, $4, $5 }'
1410 done |
1411 sort >"$_my_tickles"
1413 # Add tickles for connections that we haven't already got tickles for
1414 comm -23 "$_my_connections" "$_my_tickles" |
1415 while read _src _dst ; do
1416 ctdb addtickle $_src $_dst
1417 done
1419 # Remove tickles for connections that are no longer there
1420 comm -13 "$_my_connections" "$_my_tickles" |
1421 while read _src _dst ; do
1422 ctdb deltickle $_src $_dst
1423 done
1425 rm -f "$_my_connections" "$_my_tickles"
1426 }
1428 ########################################################
1429 # load a site local config file
1430 ########################################################
1432 [ -n "$CTDB_RC_LOCAL" -a -x "$CTDB_RC_LOCAL" ] && {
1433 . "$CTDB_RC_LOCAL"
1434 }
1436 [ -x $CTDB_BASE/rc.local ] && {
1437 . $CTDB_BASE/rc.local
1438 }
1440 [ -d $CTDB_BASE/rc.local.d ] && {
1441 for i in $CTDB_BASE/rc.local.d/* ; do
1442 [ -x "$i" ] && . "$i"
1443 done
1444 }
1446 script_name="${0##*/}" # basename
1447 service_fail_limit=1
1448 event_name="$1"