3 # This must run as root as CTDB tool commands need to access CTDB socket
4 [ "$(id -u)" -eq 0 ] || exec sudo "$0" "$@"
6 # this script needs to be installed so that statd points to it with the -H
7 # command line argument. The easiest way to do that is to put something like this in
9 # STATD_HOSTNAME="myhostname -H /etc/ctdb/statd-callout"
11 [ -n "$CTDB_BASE" ] || \
12 CTDB_BASE=$(d=$(dirname "$0") ; cd -P "$d" ; echo "$PWD")
14 . "${CTDB_BASE}/functions"
16 # Overwrite this so we get some logging
19 script_log "statd-callout" "$@"
23 load_system_config "nfs"
27 [ -n "$NFS_HOSTNAME" ] || \
28 die "NFS_HOSTNAME is not configured. statd-callout failed"
30 ############################################################
32 ctdb_setup_state_dir "service" "nfs"
34 # script_state_dir set by ctdb_setup_state_dir()
35 # shellcheck disable=SC2154
36 d="${script_state_dir}/statd-callout"
38 mkdir -p "$d" || die "Failed to create directory \"${d}\""
39 cd "$d" || die "Failed to change directory to \"${d}\""
43 ############################################################
47 _smnotify="${CTDB_HELPER_BINDIR}/smnotify"
49 # State must monotonically increase, across the entire
50 # cluster. Use seconds since epoch and hope the time is in
51 # sync across nodes. Even numbers mean service is shut down,
52 # odd numbers mean service is started.
54 # Intentionally round to an even number
55 # shellcheck disable=SC2017
56 _state_even=$(( $(date '+%s') / 2 * 2))
59 while read _sip _cip ; do
60 # NOTE: Consider optimising smnotify to read all the
61 # data from stdin and then run it in the background.
63 # Reset stateval for each serverip
64 if [ "$_sip" != "$_prev" ] ; then
65 _stateval="$_state_even"
68 # Send notifies for server shutdown
69 "$_smnotify" --client="$_cip" --ip="$_sip" \
70 --server="$_sip" --stateval="$_stateval"
71 "$_smnotify" --client="$_cip" --ip="$_sip" \
72 --server="$NFS_HOSTNAME" --stateval="$_stateval"
74 # Send notifies for server startup
75 _stateval=$((_stateval + 1))
76 "$_smnotify" --client="$_cip" --ip="$_sip" \
77 --server="$_sip" --stateval="$_stateval"
78 "$_smnotify" --client="$_cip" --ip="$_sip" \
79 --server="$NFS_HOSTNAME" --stateval="$_stateval"
85 while read _sip _cip ; do
86 _key="statd-state@${_sip}@${_cip}"
87 echo "\"${_key}\" \"\""
88 done | $CTDB ptrans "ctdb.tdb"
91 ############################################################
94 # Keep a single file to keep track of the last "add-client" or
95 # "del-client'. These get pushed to ctdb.tdb during "update",
96 # which will generally be run once each "monitor" cycle. In this
97 # way we avoid scalability problems with flood of persistent
98 # transactions after a "notify" when all the clients re-take their
102 # statd does not tell us to which IP the client connected so
103 # we must add it to all the IPs that we serve
106 # x is intentionally ignored
107 # shellcheck disable=SC2034
110 while IFS="|" read x sip node x ; do
111 [ "$node" = "$pnn" ] || continue # not us
112 key="statd-state@${sip}@${cip}"
113 echo "\"${key}\" \"${date}\"" >"$key"
118 # statd does not tell us from which IP the client disconnected
119 # so we must add it to all the IPs that we serve
121 # x is intentionally ignored
122 # shellcheck disable=SC2034
125 while IFS="|" read x sip node x ; do
126 [ "$node" = "$pnn" ] || continue # not us
127 key="statd-state@${sip}@${cip}"
128 echo "\"${key}\" \"\"" >"$key"
133 files=$(echo statd-state@*)
134 if [ "$files" = "statd-state@*" ] ; then
138 # Filter out lines for any IP addresses that are not currently
139 # hosted public IP addresses.
140 ctdb_ips=$($CTDB ip | tail -n +2)
141 sed_expr=$(echo "$ctdb_ips" |
142 awk -v pnn="$pnn" 'pnn == $2 {
143 ip = $1; gsub(/\./, "\\.", ip);
144 printf "/statd-state@%s@/p\n", ip }')
145 # Intentional multi-word expansion for multiple files
146 # shellcheck disable=SC2086
147 items=$(sed -n "$sed_expr" $files)
148 if [ -n "$items" ] ; then
149 if echo "$items" | $CTDB ptrans "ctdb.tdb" ; then
150 # shellcheck disable=SC2086
157 # we must restart the lockmanager (on all nodes) so that we get
158 # a clusterwide grace period (so other clients don't take out
159 # conflicting locks through other nodes before all locks have been
162 # we need these settings to make sure that no tcp connections survive
163 # across a very fast failover/failback
164 #echo 10 > /proc/sys/net/ipv4/tcp_fin_timeout
165 #echo 0 > /proc/sys/net/ipv4/tcp_max_tw_buckets
166 #echo 0 > /proc/sys/net/ipv4/tcp_max_orphans
168 # Delete the notification list for statd, we don't want it to
170 rm -f /var/lib/nfs/statd/sm/*
171 rm -f /var/lib/nfs/statd/sm.bak/*
173 # We must also let some time pass between stopping and
174 # restarting the lock manager. Otherwise there is a window
175 # where the lock manager will respond "strangely" immediately
176 # after restarting it, which causes clients to fail to reclaim
179 "$CTDB_NFS_CALLOUT" "stop" "nlockmgr" >/dev/null 2>&1
181 "$CTDB_NFS_CALLOUT" "start" "nlockmgr" >/dev/null 2>&1
183 # we now need to send out additional statd notifications to ensure
184 # that clients understand that the lockmanager has restarted.
185 # we have three cases:
186 # 1, clients that ignore the ip address the stat notification came from
187 # and ONLY care about the 'name' in the notify packet.
188 # these clients ONLY work with lock failover IFF that name
189 # can be resolved into an ipaddress that matches the one used
190 # to mount the share. (==linux clients)
191 # This is handled when starting lockmanager above, but those
192 # packets are sent from the "wrong" ip address, something linux
193 # clients are ok with, buth other clients will barf at.
194 # 2, Some clients only accept statd packets IFF they come from the
195 # 'correct' ip address.
196 # 2a,Send out the notification using the 'correct' ip address and also
197 # specify the 'correct' hostname in the statd packet.
198 # Some clients require both the correct source address and also the
199 # correct name. (these clients also ONLY work if the ip addresses
200 # used to map the share can be resolved into the name returned in
201 # the notify packet.)
202 # 2b,Other clients require that the source ip address of the notify
203 # packet matches the ip address used to take out the lock.
204 # I.e. that the correct source address is used.
205 # These clients also require that the statd notify packet contains
206 # the name as the ip address used when the lock was taken out.
208 # Both 2a and 2b are commonly used in lockmanagers since they maximize
209 # probability that the client will accept the statd notify packet and
210 # not just ignore it.
211 # For all IPs we serve, collect info and push to the config database
213 # Construct a sed expression to take catdb output and produce pairs of:
214 # server-IP client-IP
215 # but only for the server-IPs that are hosted on this node.
216 ctdb_all_ips=$($CTDB ip all | tail -n +2)
217 sed_expr=$(echo "$ctdb_all_ips" |
218 awk -v pnn="$pnn" 'pnn == $2 {
219 ip = $1; gsub(/\./, "\\.", ip);
220 printf "s/^key.*=.*statd-state@\\(%s\\)@\\([^\"]*\\).*/\\1 \\2/p\n", ip }')
222 statd_state=$($CTDB catdb ctdb.tdb | sed -n "$sed_expr" | sort)
223 [ -n "$statd_state" ] || exit 0
225 echo "$statd_state" | send_notifies
226 echo "$statd_state" | delete_records
228 # Remove any stale touch files (i.e. for IPs not currently
229 # hosted on this node and created since the last "update").
230 # There's nothing else we can do with them at this stage.
231 echo "$ctdb_all_ips" |
232 awk -v pnn="$pnn" 'pnn != $2 { print $1 }' |
234 rm -f "statd-state@${sip}@"*